[go: up one dir, main page]

CN102056335A - Mobile search method, device and system - Google Patents

Mobile search method, device and system Download PDF

Info

Publication number
CN102056335A
CN102056335A CN2009102208107A CN200910220810A CN102056335A CN 102056335 A CN102056335 A CN 102056335A CN 2009102208107 A CN2009102208107 A CN 2009102208107A CN 200910220810 A CN200910220810 A CN 200910220810A CN 102056335 A CN102056335 A CN 102056335A
Authority
CN
China
Prior art keywords
search
interest model
information
long
scoring
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN2009102208107A
Other languages
Chinese (zh)
Other versions
CN102056335B (en
Inventor
胡汉强
杜家春
顾翀
贾江涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Priority to CN200910220810.7A priority Critical patent/CN102056335B/en
Priority to PCT/CN2010/077773 priority patent/WO2011054245A1/en
Publication of CN102056335A publication Critical patent/CN102056335A/en
Priority to US13/464,391 priority patent/US8983971B2/en
Application granted granted Critical
Publication of CN102056335B publication Critical patent/CN102056335B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/953Querying, e.g. by the use of web search engines
    • G06F16/9535Search customisation based on user profiles and personalisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Information Transfer Between Computers (AREA)

Abstract

本发明实施例提供一种移动搜索方法、装置和系统。一个移动搜索方法包括:接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型;根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值;根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。本发明实施例,在对关键词信息进行搜索时,可以结合用户的即时兴趣模型和长期兴趣模型,从而为用户提供满足需求的、个性化的以及精确度高的搜索结果信息。

Figure 200910220810

Embodiments of the present invention provide a mobile search method, device and system. A mobile search method includes: receiving a search request, which carries keyword information to be searched and the instant interest model and long-term interest model obtained by the search application server; according to the search request, the metadata of each member engine index information and the instant interest model and long-term interest model, and calculate the correlation score value of the member engine; select one or more member engines to search the keyword information according to the correlation score value. In the embodiment of the present invention, when searching keyword information, the user's immediate interest model and long-term interest model can be combined to provide users with personalized and highly accurate search result information that meets the needs.

Figure 200910220810

Description

移动搜索方法、装置和系统 Mobile search method, device and system

技术领域technical field

本发明实施例涉及通信技术领域,尤其涉及一种移动搜索方法、装置和系统。The embodiments of the present invention relate to the field of communication technologies, and in particular, to a mobile search method, device and system.

背景技术Background technique

随着增值业务的不断发展,移动搜索业务也随之快速发展。移动搜索中一个很重要的技术亮点是精确搜索,即需要提供给用户个性化且精度较高的搜索服务。移动搜索框架是一个基于元搜索的平台,该平台整合了许多专业/垂直搜索引擎的能力,为用户提供一个全新的、综合的搜索能力。With the continuous development of value-added services, the mobile search service also develops rapidly. A very important technical highlight in mobile search is precise search, that is, it is necessary to provide users with personalized and high-precision search services. The mobile search framework is a meta-search-based platform that integrates the capabilities of many professional/vertical search engines to provide users with a brand new and comprehensive search capability.

在现有技术中,开放移动联盟(Open Mobile Alliance,以下简称:OMA)移动搜索框架包括:搜索应用服务器、搜索服务器、搜索客户端和数据源。在进行移动搜索时,搜索客户端将搜索请求发给搜索应用服务器,搜索应用服务器提取上下文信息并做查询分类,然后将搜索请求发给搜索服务器,搜索服务器将搜索请求分发给成员引擎,成员引擎在完成搜索后将搜索结果反馈给搜索服务器,搜索服务器再将搜索结果反馈给搜索应用服务器,最后搜索应用服务器将最终的搜索结果返回给搜索客户端。In the prior art, an Open Mobile Alliance (OMA) mobile search framework includes: a search application server, a search server, a search client and a data source. When performing mobile search, the search client sends the search request to the search application server, the search application server extracts the context information and classifies the query, and then sends the search request to the search server, and the search server distributes the search request to the member engines, and the member engines After the search is completed, the search result is fed back to the search server, and then the search server feeds back the search result to the search application server, and finally the search application server returns the final search result to the search client.

在实现本发明过程中,发明人发现现有技术中至少存在如下问题:现有基于OMA移动搜索框架的移动搜索,其个性化程度不高,无法满足用户对搜索精度的需求。In the process of realizing the present invention, the inventors found that there are at least the following problems in the prior art: the existing mobile search based on the OMA mobile search framework is not highly personalized and cannot meet the user's demand for search accuracy.

发明内容Contents of the invention

本发明实施例提供一种移动搜索方法、装置和系统。Embodiments of the present invention provide a mobile search method, device and system.

本发明实施例提供一种移动搜索方法,包括:An embodiment of the present invention provides a mobile search method, including:

接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型;Receiving a search request, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server;

根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值;According to the search request, the meta index information of each member engine, and the instant interest model and long-term interest model, calculate the relevance score value of the member engine;

根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。Selecting one or more member engines to search the keyword information according to the correlation score value.

本发明实施例另提供一种移动搜索方法,包括:An embodiment of the present invention further provides a mobile search method, including:

向一个或多个成员引擎发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;Send a search request to one or more member engines, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server;

接收所述一个或多个成员引擎根据所述关键词信息、所述即时兴趣模型和长期兴趣模型获取的搜索结果信息以及与所述搜索结果信息对应的评分信息;receiving search result information obtained by the one or more member engines according to the keyword information, the immediate interest model and the long-term interest model, and scoring information corresponding to the search result information;

根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息发送给所述搜索应用服务器。Re-scoring and sorting the search result information according to the scoring information and relevant factor information, obtaining the re-scoring and sorting search result information, and sending the re-scoring and sorting search result information to the search application server.

本发明实施例再提供一种移动搜索方法,包括:An embodiment of the present invention further provides a mobile search method, including:

接收搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;Receive the search request sent by the search application server, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server;

接收成员引擎根据所述关键词信息搜索获取的搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;receiving the search result information obtained by the member engine according to the keyword information search, and performing scoring and sorting processing on the search result information according to the instant interest model and the long-term interest model;

将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述搜索应用服务器。Send the search result information after scoring and sorting processing and the corresponding scoring information to the search application server.

本发明实施例又提供一种移动搜索方法,包括:An embodiment of the present invention further provides a mobile search method, including:

接收搜索请求,所述搜索请求中携带即时兴趣模型和长期兴趣模型;Receiving a search request, the search request carries an immediate interest model and a long-term interest model;

根据所述搜索请求进行搜索获取搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;performing a search according to the search request to obtain search result information, and performing scoring and ranking processing on the search result information according to the immediate interest model and the long-term interest model;

返回评分排序处理后的搜索结果信息。Returns the search result information after scoring and sorting.

本发明实施例还提供一种移动搜索方法,包括:The embodiment of the present invention also provides a mobile search method, including:

接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;receiving a search request message sent by a search client, wherein the search request message carries keyword information;

从用户数据库中提取即时兴趣模型和长期兴趣模型;Extract immediate and long-term interest models from the user database;

向搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型,以使所述搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。Sending a search request to a search server, the search request carrying keyword information and the immediate interest model and long-term interest model, so that the search server can perform a search on the keyword information according to the immediate interest model and the long-term interest model search.

本发明实施例提供一种搜索服务器,包括:An embodiment of the present invention provides a search server, including:

第一接收模块,用于接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型;The first receiving module is configured to receive a search request, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server;

第一处理模块,用于根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值;A first processing module, configured to calculate the relevance score value of the member engine according to the search request, the meta index information of each member engine, and the immediate interest model and long-term interest model;

第一搜索模块,用于根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。A first search module, configured to select one or more member engines to search the keyword information according to the correlation score.

本发明实施例另提供一种搜索服务器,包括:An embodiment of the present invention further provides a search server, including:

第二发送模块,用于向一个或多个成员引擎发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;The second sending module is used to send a search request to one or more member engines, and the search request carries the keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server;

第二接收模块,用于接收所述一个或多个成员引擎根据所述关键词信息、所述即时兴趣模型和长期兴趣模型获取的搜索结果信息以及与所述搜索结果信息对应的评分信息;A second receiving module, configured to receive search result information obtained by the one or more member engines according to the keyword information, the immediate interest model and the long-term interest model, and scoring information corresponding to the search result information;

第二处理模块,用于根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息通过所述第二发送模块发送给所述搜索应用服务器。The second processing module is used to re-score and sort the search result information according to the scoring information and related factor information, obtain the re-scoring and sorting search result information, and pass the re-scoring and sorting search result information through The second sending module sends to the search application server.

本发明实施例再提供一种搜索服务器,包括:An embodiment of the present invention provides a search server, including:

第三接收模块,用于接收搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;并接收成员引擎根据所述关键词信息搜索获取的搜索结果信息;The third receiving module is used to receive the search request sent by the search application server, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server; Search result information obtained by keyword information search;

第三处理模块,用于根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;A third processing module, configured to perform scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model;

第三发送模块,用于将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述搜索应用服务器。The third sending module is configured to send the search result information after scoring and sorting processing and corresponding scoring information to the search application server.

本发明实施例提供一种成员引擎设备,包括:An embodiment of the present invention provides a member engine device, including:

第四接收模块,用于接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及即时兴趣模型和长期兴趣模型;The fourth receiving module is used to receive a search request, and the search request carries keyword information to be searched and an immediate interest model and a long-term interest model;

第四处理模块,用于根据所述搜索请求对所述关键词信息进行搜索获取搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;A fourth processing module, configured to search the keyword information according to the search request to obtain search result information, and perform scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model;

第四发送模块,用于返回评分排序处理后的搜索结果信息。The fourth sending module is used to return the search result information after scoring and sorting.

本发明实施例提供一种搜索应用服务器,包括:An embodiment of the present invention provides a search application server, including:

第五接收模块,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;The fifth receiving module is configured to receive a search request message sent by a search client, where the search request message carries keyword information;

第五处理模块,用于从用户数据库中提取即时兴趣模型和长期兴趣模型;The fifth processing module is used to extract an immediate interest model and a long-term interest model from the user database;

第五发送模块,用于向搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型,以使所述搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。The fifth sending module is configured to send a search request to a search server, wherein the search request carries keyword information and the immediate interest model and the long-term interest model, so that the search server can search according to the immediate interest model and the long-term interest model The keyword information is searched.

本发明实施例提供一种移动搜索系统,包括:An embodiment of the present invention provides a mobile search system, including:

第一搜索应用服务器,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第一搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;The first search application server is configured to receive a search request message sent by a search client, wherein the search request message carries keyword information; extract an immediate interest model and a long-term interest model from a user database; send a search request to the first search server , the search request carries keyword information and the immediate interest model and long-term interest model;

第一搜索服务器,用于接收所述第一搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述第一搜索应用服务器获取的即时兴趣模型和长期兴趣模型;根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值,根据所述相关度评分值从第一成员引擎设备中选择一个或多个成员引擎对所述关键词信息进行搜索;The first search server is configured to receive the search request sent by the first search application server, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the first search application server ; According to the search request, the meta-index information of each member engine and the instant interest model and long-term interest model, calculate the relevance score value of the member engine, according to the correlation score value from the first member engine device Selecting one or more member engines to search for the keyword information;

第一成员引擎设备,用于接收所述第一搜索服务器发送的搜索请求,对所述关键词信息进行搜索,并将搜索结果信息发送给所述第一搜索服务器,以使所述第一搜索服务器通过所述第一搜索应用服务器将所述搜索结果信息反馈给所述搜索客户端。The first member engine device is configured to receive a search request sent by the first search server, search for the keyword information, and send search result information to the first search server, so that the first search The server feeds back the search result information to the search client through the first search application server.

本发明实施例另提供一种移动搜索系统,包括:An embodiment of the present invention further provides a mobile search system, including:

第二搜索应用服务器,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第二搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;The second search application server is configured to receive the search request message sent by the search client, the search request message carries keyword information; extract the immediate interest model and the long-term interest model from the user database; send a search request to the second search server , the search request carries keyword information and the immediate interest model and long-term interest model;

第二搜索服务器,用于向第二成员引擎设备发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及第二搜索应用服务器获取的即时兴趣模型和长期兴趣模型;接收所述第二成员引擎设备反馈的搜索结果信息以及与所述搜索结果信息对应的评分信息;根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息发送给所述第二搜索应用服务器;The second search server is configured to send a search request to the second member engine device, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the second search application server; receiving the first The search result information fed back by the two-member engine device and the scoring information corresponding to the search result information; re-scoring and sorting the search result information according to the scoring information and related factor information, and obtaining the re-scoring and sorting search result information , and sending the re-scored and sorted search result information to the second search application server;

第二成员引擎设备,用于接收所述第二搜索服务器发送的搜索请求,对所述关键词信息进行搜索,根据所述即时兴趣模型和长期兴趣模型获取搜索结果信息以及与所述搜索结果信息对应的评分信息,并将所述搜索结果信息以及评分信息发送给所述第二搜索服务器。The second member engine device is configured to receive a search request sent by the second search server, search for the keyword information, obtain search result information and match the search result information according to the immediate interest model and the long-term interest model Corresponding scoring information, and sending the search result information and scoring information to the second search server.

本发明实施例再提供一种移动搜索系统,包括:An embodiment of the present invention further provides a mobile search system, including:

第三搜索应用服务器,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第三搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;The third search application server is configured to receive the search request message sent by the search client, the search request message carries keyword information; extract the immediate interest model and the long-term interest model from the user database; send a search request to the third search server , the search request carries keyword information and the immediate interest model and long-term interest model;

第三搜索服务器,用于接收第三搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;接收第三成员引擎设备根据所述关键词信息搜索获取的搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;并将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述第三搜索应用服务器;The third search server is used to receive the search request sent by the third search application server, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server; receiving the third member engine The device searches the search result information acquired according to the keyword information, and performs scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model; sorts the search result information after scoring and corresponding scoring sending the information to the third search application server;

第三成员引擎设备,用于接收所述第三搜索服务器发送的搜索请求,对所述关键词信息进行搜索,获取所述搜索结果信息。The third member engine device is configured to receive the search request sent by the third search server, search the keyword information, and obtain the search result information.

本发明实施例,在对关键词信息进行搜索时,可以结合用户的即时兴趣模型和长期兴趣模型,从而为用户提供满足需求的、个性化的以及精确度高的搜索结果信息。In the embodiment of the present invention, when searching keyword information, the user's immediate interest model and long-term interest model can be combined to provide users with personalized and highly accurate search result information that meets the needs.

附图说明Description of drawings

为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍,显而易见地,下面描述中的附图是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings without any creative effort.

图1为本发明移动搜索方法实施例一的流程图;FIG. 1 is a flow chart of Embodiment 1 of the mobile search method of the present invention;

图2为本发明移动搜索方法实施例二的信令流程图;FIG. 2 is a signaling flow chart of Embodiment 2 of the mobile search method of the present invention;

图3为本发明移动搜索方法实施例三的流程图;FIG. 3 is a flow chart of Embodiment 3 of the mobile search method of the present invention;

图4为本发明移动搜索方法实施例四的流程图;FIG. 4 is a flow chart of Embodiment 4 of the mobile search method of the present invention;

图5为本发明移动搜索方法实施例五的信令流程图;FIG. 5 is a signaling flow chart of Embodiment 5 of the mobile search method of the present invention;

图6为本发明移动搜索方法实施例六的流程图;FIG. 6 is a flow chart of Embodiment 6 of the mobile search method of the present invention;

图7为本发明移动搜索方法实施例七的信令流程图;FIG. 7 is a signaling flow chart of Embodiment 7 of the mobile search method of the present invention;

图8为本发明移动搜索方法实施例八的流程图;FIG. 8 is a flow chart of Embodiment 8 of the mobile search method of the present invention;

图9为本发明搜索服务器实施例一的结构示意图;FIG. 9 is a schematic structural diagram of Embodiment 1 of the search server of the present invention;

图10为本发明搜索服务器实施例二的结构示意图;FIG. 10 is a schematic structural diagram of Embodiment 2 of the search server of the present invention;

图11为本发明搜索服务器实施例三的结构示意图;FIG. 11 is a schematic structural diagram of Embodiment 3 of the search server of the present invention;

图12为本发明搜索服务器实施例四的结构示意图;FIG. 12 is a schematic structural diagram of Embodiment 4 of the search server of the present invention;

图13为本发明搜索服务器实施例五的结构示意图;FIG. 13 is a schematic structural diagram of Embodiment 5 of the search server of the present invention;

图14为本发明成员引擎设备实施例一的结构示意图;FIG. 14 is a schematic structural diagram of Embodiment 1 of the member engine device of the present invention;

图15为本发明成员引擎设备实施例二的结构示意图;FIG. 15 is a schematic structural diagram of Embodiment 2 of the member engine device of the present invention;

图16为本发明搜索应用服务器实施例的结构示意图;Fig. 16 is a schematic structural diagram of an embodiment of a search application server in the present invention;

图17为本发明移动搜索系统实施例一的结构示意图;FIG. 17 is a schematic structural diagram of Embodiment 1 of the mobile search system of the present invention;

图18为本发明移动搜索系统实施例二的结构示意图;FIG. 18 is a schematic structural diagram of Embodiment 2 of the mobile search system of the present invention;

图19为本发明移动搜索系统实施例三的结构示意图。FIG. 19 is a schematic structural diagram of Embodiment 3 of the mobile search system of the present invention.

具体实施方式Detailed ways

为使本发明实施例的目的、技术方案和优点更加清楚,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

图1为本发明移动搜索方法实施例一的流程图,如图1所示,本实施例的方法可以包括:Fig. 1 is a flowchart of Embodiment 1 of the mobile search method of the present invention. As shown in Fig. 1, the method of this embodiment may include:

步骤101、接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型。Step 101: Receive a search request, the search request carries keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server.

举例来说,搜索应用服务器可以根据搜索客户端发送的搜索请求消息,从用户数据库中获取用户的即时兴趣模型和长期兴趣模型。例如,从用户数据库,如用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前查询q(t)处于同一搜索上下文会话(Search Context Session)的查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型,所述搜索上下文会话为当前查询q(t)发生的前一段预设的时间,如半个小时,包括q(t)当前发生的时间。在提取即时兴趣模型和长期兴趣模型后,搜索应用服务器即可向搜索服务器发送搜索请求,从而使得搜索服务器可以根据该搜索请求中携带的用户的即时兴趣模型和长期兴趣模型对所需搜索的关键字信息进行搜索。For example, the search application server can obtain the user's immediate interest model and long-term interest model from the user database according to the search request message sent by the search client. For example, extract the user's long-term interest model from the user database, such as the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract from the current query q( t) Extract the user's instant interest model from the relevant data of the query sequence q(1), ..., q(t-1), q(t) in the same search context session (Search Context Session), the search context A session is a preset period of time before the occurrence of the current query q(t), such as half an hour, including the current occurrence time of q(t). After extracting the immediate interest model and the long-term interest model, the search application server can send a search request to the search server, so that the search server can determine the key information of the desired search according to the user's immediate interest model and long-term interest model carried in the search request. word information to search.

用户的兴趣模型可以用n个维度来表示如:新闻、体育、娱乐、财经、科技、房产、游戏、女性、论坛、天气、商品、家电、音乐、读书、博客、手机、军事、教育、旅游、彩信、彩铃、餐饮、民航、工业、农业、电脑、地理等。用户对每个维度的兴趣的评分值所组成的一个向量W(r1,r2,r3,......,rn)则为用户的兴趣模型。The user's interest model can be represented by n dimensions such as: news, sports, entertainment, finance, technology, real estate, games, women, forums, weather, commodities, home appliances, music, reading, blogs, mobile phones, military, education, tourism , MMS, CRBT, catering, civil aviation, industry, agriculture, computer, geography, etc. A vector W(r1, r2, r3, . . . , rn) composed of rating values of the user's interest in each dimension is the user's interest model.

如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由用户的所有搜索历史数据和用户的静态档案profile计算得到,则兴趣模型W(r1,r2,r3,...rn)为用户的长期兴趣模型。如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到的,那么W(r1,r2,r3,...rn)为用户的即时兴趣模型,在此需要说明的是本发明实施例中的查询q(t)与搜索请求相对应,即每个搜索请求都会有一个查询q(t)。If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is calculated from all the user's search history data and the user's static profile profile, then the interest model W(r1, r2 , r3,...rn) is the user's long-term interest model. If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is determined by the query sequence q(1) in the same search context session as the current query q(t), ... , q(t-1), calculated from the related data of q(t), then W(r1, r2, r3,...rn) is the user's instant interest model, and what needs to be explained here is the embodiment of the present invention The query q(t) in corresponds to the search request, that is, each search request will have a query q(t).

步骤102、根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值。Step 102 , according to the search request, the meta index information of each member engine, and the instant interest model and long-term interest model, calculate the relevance score value of the member engine.

成员引擎是具体用于搜索关键词信息的服务器,成员引擎可以向搜索服务器上报其元索引信息。具体的,成员引擎的元索引信息为关于该成员引擎对应的数据库,子数据库,数据库或者子数据库中包含的文档或者记录,以及,该文档或者记录中包含的术语的统计数据。The member engine is a server specifically used for searching keyword information, and the member engine can report its meta-index information to the search server. Specifically, the meta-index information of the member engine is statistical data about the database, sub-database, documents or records contained in the database or sub-database corresponding to the member engine, and the terms contained in the documents or records.

元索引信息可以包括下述信息之一或者其任意组合:Meta index information can include one or any combination of the following information:

术语最大归一化权重向量mnw=(mnw1,mnw2,...,mnwi,...mnwp),其中mnwi为术语ti相对于所述成员引擎对应的数据库或者子数据库中的所有文档的最大归一化权重;Term maximum normalized weight vector mnw=(mnw1, mnw2, . unified weight;

术语平均归一化权重向量anw=(anw1,anw2,...,anwi......,anwp),其中anwi为术语ti相对于所述成员引擎对应的数据库或者子数据库中的所有文档的平均归一化权重;Term average normalized weight vector anw=(anw1, anw2, . . . , anwi . The average normalized weight of ;

数据库或者子数据库中的文档的兴趣模型最大归一化权重向量mnv=(mnv1,mnv2,......,mnvi,......,mnvn),其中mnvi为所述文档的兴趣模型的第i个维度相对于所述成员引擎对应的数据库或者子数据库中的所有文档的最大归一化权重;The maximum normalized weight vector mnv=(mnv1, mnv2, ..., mnvi, ..., mnvn) of the interest model of the document in the database or sub-database, wherein mnvi is the interest of the document The i-th dimension of the model is relative to the maximum normalized weight of all documents in the database or sub-database corresponding to the member engine;

数据库或者子数据库中的文档的兴趣模型平均归一化权重向量anv=(anv1,anv2,......,anvi,......,anvn),其中anvi为文档的兴趣模型的第i个维度相对于所述成员引擎对应的数据库或者子数据库中的所有文档的平均归一化权重;The average normalized weight vector anv=(anv1, anv2, ..., anvi, ..., anvn) of the interest model of the document in the database or the sub-database, where anvi is the interest model of the document The i-th dimension is relative to the average normalized weight of all documents in the database or sub-database corresponding to the member engine;

术语ti相对于该数据库的全局反向文档频率gidfi,其中gidfi=1/dfi,dfi为该元索引对应的数据库中包含术语ti的文档的数量;The term ti is relative to the global inverse document frequency gidfi of the database, where gidfi=1/dfi, and dfi is the number of documents containing the term ti in the database corresponding to the meta index;

文档的兴趣模型第i个维度对应的全局反向文档频率IM_gidfi,其中IM_gidfi=1/IM_IDFi,IM_IDFi为数据库或者子数据库中包含的属于文档的兴趣模型的第i个维度的术语的文档的个数;The global reverse document frequency IM_gidfi corresponding to the i-th dimension of the document’s interest model, where IM_gidfi=1/IM_IDFi, IM_IDFi is the number of documents contained in the database or sub-database that belong to the i-th dimension of the document’s interest model ;

术语ti相对于该数据库的全局反向文档频率gidfi=log(n/(gdfi+1)),其中gdfi为所有成员引擎对应数据库或者子数据库中包含术语ti的文档的数量的总和,n为所有成员引擎所包含的所有文档数量的总和;或者,The term ti is relative to the global reverse document frequency of the database gidfi=log(n/(gdfi+1)), where gdfi is the sum of the number of documents containing the term ti in the corresponding database or sub-database of all member engines, and n is all the sum of all document counts contained in the member engine; or,

文档的兴趣模型第i个维度对应的全局反向文档频率IM_gidfi=log(n/(IM_gdfi+1)),IM_gdfi为所有成员引擎对应的数据库或子数据库中包含属于文档的兴趣模型的第i个维度的术语的文档个数的总和,n为所有成员引擎所包含的所有文档数量的总和。The global inverse document frequency IM_gidfi=log(n/(IM_gdfi+1)) corresponding to the i-th dimension of the interest model of the document, IM_gdfi is the i-th one of the interest model belonging to the document in the database or sub-database corresponding to all member engines The sum of the number of documents of the dimension term, n is the sum of the number of all documents contained in all member engines.

搜索服务器可以根据各成员引擎上报的元索引信息以及所述即时兴趣模型和长期兴趣模型,获取各成员引擎的选择结果信息。举例来说,搜索服务器可以将元索引信息与即时兴趣模型、长期兴趣模型以及查询请求分别进行匹配处理,并将选择匹配较好的成员引擎作为后续搜索关键词信息所使用的成员引擎,该选择结果信息可以为选择出的成员引擎的ID信息。选择出的成员引擎可以与用户的长期兴趣、即时兴趣以及所需搜索的关键词信息进行较好的匹配,从而使得搜索服务器可以将所需搜索的关键词交付给选择出的成员引擎进行搜索,从而获取较为精确的搜索结果信息。The search server may acquire the selection result information of each member engine according to the meta index information reported by each member engine and the immediate interest model and long-term interest model. For example, the search server can match the meta-index information with the instant interest model, the long-term interest model, and the query request, and select the member engine that matches better as the member engine used for subsequent search keyword information. The result information may be the ID information of the selected member engine. The selected member engine can better match the user's long-term interest, instant interest and keyword information to be searched, so that the search server can deliver the keyword to be searched to the selected member engine for search, In order to obtain more accurate search result information.

步骤103、根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。Step 103: Select one or more member engines to search the keyword information according to the correlation score.

搜索服务器在从各成员引擎中选择出所需使用的成员引擎后,即可向选择出的成员引擎发送搜索请求。After the search server selects the required member engine from among the member engines, it can send a search request to the selected member engine.

本实施例中,搜索服务器在请求成员引擎搜索关键词信息之前,可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎进行选择,从而能够选择获取与关键词信息、即时兴趣模型和长期兴趣模型匹配较好的成员引擎对该关键词信息进行搜索,从而能够获取精度较高的搜索结果信息,进一步满足用户的搜索需求。In this embodiment, before the search server requests the member engine to search for keyword information, it can select the member engine according to the user's immediate interest model and long-term interest model, so as to be able to select and acquire keyword information, immediate interest model and long-term interest model The member engine with better matching searches for the keyword information, so as to obtain search result information with higher precision, and further satisfy the user's search needs.

在本发明移动搜索方法另一个实施例中,所述即时兴趣模型可以为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。In another embodiment of the mobile search method of the present invention, the instant interest model may be an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by the same search context session as the current query q(t). The relevant data of the query sequence q(1),...,q(t-1), q(t) is calculated, and the search context session includes the current query q(t) before the occurrence of the current query q(t) A preset period of time including the time of occurrence.

具体来说,本实施例可以应用条件随机场(Conditional Random Field,以下简称:CRF)模型计算在给定与当前查询q(t)处于同一搜索上下文会话(Search Context Session)的查询序列q(1),....q(t-1),q(t)的条件下,当前查询q(t)的输出类型的条件概率,把该条件概率值作为即时兴趣模型与该输出类型对应的兴趣维度的评分值。Specifically, this embodiment can apply a conditional random field (Conditional Random Field, hereinafter referred to as: CRF) model to calculate the query sequence q(1 ),....q(t-1), q(t), the conditional probability of the output type of the current query q(t), and the conditional probability value is used as the interest corresponding to the output type of the instant interest model The score value of the dimension.

举例来说,本实施例可以定义G=(V,E)为一个无向图,Y={Yv|v∈V}.即V中的每个节点对应一个随机变量所表示的标记序列的成分Yv,如果每个随机量Yv对于G遵守马尔可夫属性,那么(X,Y)就是一个条件随机场,而且在给定X和所有其他随机变量Y{u|u≠v,{u,v}∈V}的条件下,随机变量Yv的概率P(Yv|X,Yu,u≠v,{u.v}∈V)即等于P(Yv|X,Yu,(u,v)∈E)。For example, this embodiment can define G=(V, E) as an undirected graph, Y={Y v |v∈V}. That is, each node in V corresponds to a sequence of labels represented by a random variable component Y v , if each random quantity Y v obeys the Markov property for G, then (X, Y) is a conditional random field, and given X and all other random variables Y {u|u≠v, { Under the condition of u, v}∈V} , the probability P(Y v |X, Y u , u≠v, {uv}∈V) of random variable Y v is equal to P(Y v |X, Y u , ( u, v) ∈ E).

根据马尔可夫属性和最大熵的原理,可以推导出条件随机场的经典条件概率公式:According to the Markov property and the principle of maximum entropy, the classical conditional probability formula of the conditional random field can be derived:

给定观察序列x的前提下,观察序列的状态标记序列y的条件概率等于:Given the observation sequence x, the conditional probability of the state label sequence y of the observation sequence is equal to:

PP θθ (( ythe y || xx )) == 11 ZZ (( xx )) expexp (( ΣΣ ee ∈∈ EE. ,, kk λλ kk ff kk (( ee ,, ythe y || ee ,, xx )) ++ ΣΣ vv ∈∈ VV ,, kk uu kk gg kk (( vv ,, ythe y || vv ,, xx )) ))

其中,x为观察序列,y为标记状态序列,y|S为与子图S的顶点相关联的序列y的成分的集合,f,g为特征函数,λ,μ为特征函数的权重值,Z(x)为归一化因子。Among them, x is the observation sequence, y is the marked state sequence, y| S is the set of components of the sequence y associated with the vertices of the subgraph S, f, g are the characteristic functions, λ, μ are the weight values of the characteristic functions, Z(x) is a normalization factor.

给定处于同一session的查询序列q=q1,...,q(T-1),q(T),输出查询序列对应的类型序列C=c1,...cT-1,cT的条件概率:Given a query sequence q=q1,...,q(T-1), q(T) in the same session, output the type sequence C=c 1 ,...c T-1 , c T corresponding to the query sequence The conditional probability of :

令ci的取值空间为|C|,令c0=start,cT+1=end,将状态start、end加入|C|Let the value space of c i be |C|, let c 0 = start, c T+1 = end, add the states start and end to |C|

pp (( cc || qq )) == 11 ZZ (( qq )) ΠΠ tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) ,,

其中: Z ( q ) = Σ c Π t = 1 T + 1 M t ( c t - 1 , c t | q ) , 为归一化因子。in: Z ( q ) = Σ c Π t = 1 T + 1 m t ( c t - 1 , c t | q ) , is the normalization factor.

Mm tt (( cc tt -- 11 ,, cc tt || qq )) == expexp (( ΣΣ kk λλ kk ff kk (( cc tt -- 11 ,, cc tt ,, qq )) ++ ΣΣ kk uu kk gg kk (( cc tt ,, qq )) ))

定义一个|C|×|C|的矩阵:Define a |C|×|C| matrix:

Mt(q)=[Mt(ct-1,ct|q)]M t (q)=[M t (c t-1 , c t |q)]

那么Z(q)等于M1(q)*...MT(q)*MT+1(q)矩阵的(start,end)项。Then Z(q) is equal to the (start, end) items of the M 1 (q)*...M T (q)*M T+1 (q) matrix.

ZZ (( qq )) == ΣΣ cc ΠΠ tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) == ΣΣ startcstartc 11 .. .. .. cc TT -- 11 cc TT endend ΠΠ tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq ))

== (( ΠΠ tt == 11 TT ++ 11 Mm tt (( qq )) )) startstart ,, endend

令θ=(λ1,.λ2,......;u1,.u2,......)Let θ=(λ 1 , .λ 2 , . . . ; u 1 , .u 2 , . . . )

参数θ的确定:Determination of parameter θ:

给定训练数据

Figure B2009102208107D0000124
和经验分布
Figure B2009102208107D0000125
given training data
Figure B2009102208107D0000124
and experience distribution
Figure B2009102208107D0000125

训练数据的对数似然函数为:The log-likelihood function of the training data is:

LL (( θθ )) == ΣΣ ii == 11 NN loglog PP θθ (( cc (( ii )) || qq (( ii )) ))

∝∝ ΣΣ qq ,, cc pp ~~ (( qq ,, cc )) loglog (( pp θθ (( cc || qq )) ))

求θ使得L(θ)取得最大值。Find θ such that L(θ) takes the maximum value.

本实施例可以用通用迭代算法(General Iterative Scaling,以下简称:GIS)算法求θ:In this embodiment, the general iterative algorithm (General Iterative Scaling, hereinafter referred to as: GIS) algorithm can be used to find θ:

(a)求Efk、Egk(a) Find Ef k , Eg k :

EfEf kk == ΣΣ qq ,, cc pp θθ (( qq ,, cc )) ff kk (( qq ,, cc ))

== ΣΣ qq ,, cc pp θθ (( qq )) pp θθ (( cc || qq )) ff kk (( qq ,, cc ))

≈≈ ΣΣ qq ,, cc pp ~~ (( qq )) pp θθ (( cc || qq )) ff kk (( qq ,, cc ))

== ΣΣ qq pp ~~ (( qq )) ΣΣ ii == 11 TT ΣΣ cc ii -- 11 cc ii (( pp θθ (( cc ii -- 11 ,, cc ii || qq )) ff kk (( cc ii -- 11 ,, cc ii ,, qq )) ))

pp θθ (( cc ii -- 11 ,, cc ii || qq )) == ΣΣ startcstartc 11 .. .. .. cc ii -- 22 cc ii ++ 11 .. .. .. cc TendTend pp θθ (( startcstartc 11 ,, .. .. .. ,, cc TT endend || qq ))

== 11 ZZ (( qq )) ΣΣ startcstartc 11 .. .. .. cc ii -- 22 cc ii ++ 11 .. .. .. cc TendTend ΠΠ tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq ))

== 11 ZZ (( qq )) (( ΣΣ startcstartc 11 .. .. .. cc ii -- 22 ΠΠ tt == 11 ii -- 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) )) Mm ii (( cc ii -- 11 ,, cc ii || qq )) (( ΣΣ CC ii ++ 11 .. .. .. CC TT endend ΠΠ tt == ii ++ 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) ))

== 11 ZZ (( qq )) (( (( ΠΠ tt == 11 ii -- 11 Mm tt (( qq )) )) )) startstart ,, cc ii -- 11 Mm ii (( cc ii -- 11 ,, cc ii || qq )) (( ΠΠ tt == ii ++ 11 TT ++ 11 Mm tt (( qq )) )) cc ii ,, endend ))

== 11 ZZ (( qq )) αα ii -- 11 (( cc ii -- 11 || qq )) Mm ii (( cc ii -- 11 ,, cc ii || qq )) ββ ii (( cc ii || qq )) ))

其中:in:

αi(q)为1×|C|向量,α i (q) is a 1×|C| vector,

αi(q)=αi-1Mi(q)α i (q) = α i-1 M i (q)

如果ct=start,α0(ct|q)=1If c t =start, α 0 (c t |q)=1

否则α0(ct|q)=0Otherwise α 0 (c t |q)=0

β(i)为1x|C|向量,β(i) is a 1x|C| vector,

β(i)T=Mi+1(q)β(i+1)T β(i) T =M i+1 (q)β(i+1) T

如果ct=endβT+1(ct|q)=1If c t =endβ T+1 (c t |q)=1

否则,βT+1(ct|q)=0Otherwise, β T+1 (c t |q)=0

Egeg kk == ΣΣ qq ,, cc pp θθ (( qq ,, cc )) gg kk (( qq ,, cc ))

== ΣΣ qq ,, cc pp θθ (( qq )) pp θθ (( cc || qq )) gg kk (( qq ,, cc ))

≈≈ ΣΣ qq ,, cc pp ~~ (( qq )) pp θθ (( cc || qq )) gg kk (( qq ,, cc ))

== ΣΣ qq pp ~~ (( qq )) ΣΣ ii == 11 TT ΣΣ cc ii pp θθ (( cc ii || qq )) gg kk (( cc ii ,, qq ))

pp θθ (( cc ii || qq )) == ΣΣ cc 11 .. .. .. cc ii -- 11 cc ii ++ 11 .. .. .. cc TT pp θθ (( cc 11 ,, .. .. .. ,, cc TT || qq ))

== 11 ZZ (( qq )) ΣΣ cc 11 .. .. .. cc ii -- 11 cc ii ++ 11 .. .. .. cc TT ΠΠ tt == 11 TT Mm tt (( cc tt -- 11 ,, cc tt ,, qq ))

== 11 ZZ (( qq )) (( ΣΣ cc 11 .. .. .. cc ii -- 11 ΠΠ tt == 11 ii Mm tt (( cc tt -- 11 ,, cc tt ,, qq )) )) (( ΣΣ cc ii ++ 11 .. .. .. cc TT ΠΠ tt == ii ++ 11 TT Mm tt (( cc tt -- 11 ,, cc tt ,, qq )) ))

== 11 ZZ (( qq )) (( (( ΠΠ tt == 11 ii Mm tt (( qq )) )) )) startstart ,, cc ii (( ΠΠ tt == ii ++ 11 TT ++ 11 Mm tt (( qq )) )) cc ii ,, endend ))

== 11 ZZ (( qq )) αα (( ii )) ββ (( ii ))

(b)求

Figure B2009102208107D0000144
(b) seeking
Figure B2009102208107D0000144

EE. ~~ ff kk == ΣΣ qq ,, cc pp ~~ (( qq ,, cc )) ff kk (( qq ,, cc ))

== ΣΣ qq ,, cc pp ~~ (( qq )) pp ~~ (( cc || qq )) ff kk (( qq ,, cc ))

== ΣΣ qq pp ~~ (( qq )) ΣΣ ii == 11 TT ΣΣ cc ii -- 11 cc ii pp ~~ (( cc ii -- 11 ,, cc ii || qq )) ff kk (( cc ii -- 11 ,, cc ii ,, qq ))

EE. ~~ gg kk == ΣΣ qq ,, cc pp ~~ (( qq ,, cc )) gg kk (( qq ,, cc ))

== ΣΣ qq ,, cc pp ~~ (( qq )) pp ~~ (( cc || qq )) gg kk (( qq ,, cc ))

== ΣΣ qq pp ~~ (( qq )) ΣΣ ii == 11 TT ΣΣ cc ii pp ~~ (( cc ii || qq )) gg kk (( cc ii ,, qq ))

(c)求迭代求λk、uk,直到λk、uk收敛:(c) Calculate λ k and u k iteratively until λ k and u k converge:

Figure B2009102208107D00001411
其中S1为大于1的常数,使得对任何
Figure B2009102208107D00001412
Figure B2009102208107D00001411
where S 1 is a constant greater than 1, such that for any
Figure B2009102208107D00001412

λk+1=λk+δλk λ k+1 =λ k +δλ k

Figure B2009102208107D00001413
其中S2为大于1的常数,使得对任何q、c, Σ k = 0 n g k ( q , c ) = S 2
Figure B2009102208107D00001413
where S 2 is a constant greater than 1, so that for any q, c, Σ k = 0 no g k ( q , c ) = S 2

uk+1=uk+δuk u k+1 =u k +δu k

重复(a)、(b)、(c)步骤直到λk、uk收敛。Repeat steps (a), (b) and (c) until λ k and u k converge.

给定处于同一session查询序列q=q1,...,q(T-1),q(T),当前查询q(T)属于类型cT的条件概率:Given the query sequence q=q1,...,q(T-1), q(T) in the same session, the conditional probability that the current query q(T) belongs to type c T :

pp (( cc TT || qq )) == ΣΣ cc 11 .. .. .. cc TT -- 11 pp (( cc || qq ))

== 11 ZZ (( qq )) (( ΣΣ startcstartc 11 .. .. .. cc TT -- 11 ΠΠ tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) ))

== 11 ZZ (( qq )) (( ΣΣ startcstartc 11 .. .. .. cc TT -- 11 ΠΠ tt == 11 TT Mm tt (( cc tt -- 11 ,, cc tt || qq )) )) Mm TT ++ 11 (( cc TT ,, cc endend || qq )) ))

== 11 ZZ (( qq )) (( (( ΠΠ tt == 11 TT Mm tt (( qq )) )) )) startstart ,, cc TT Mm TT ++ 11 (( cc TT ,, cc endend || qq ))

== 11 ZZ (( qq )) αα TT (( cc TT || qq )) Mm TT ++ 11 (( cc TT ,, cc endend || qq ))

把p(cT|q)作为即时兴趣模型类型为cT的对应维度的评分值。Take p(c T |q) as the score value of the corresponding dimension of the instant interest model type c T.

本地特征函数gk的选取:Selection of local feature function g k :

(1)给每个领域类型cT的所有主题词和相关词赋予一定的权重,由这些主题词和相关词的权重组成一个领域cT的向量,cT(t1,...,tn-1,tn)(1) Give certain weights to all subject words and related words of each field type c T , and form a vector of field c T by the weights of these subject words and related words, c T (t 1 ,...,t n-1 , t n )

cT中的词的权重的分配方法有两种,There are two ways to assign the weight of words in c T ,

一种是人工分配权重的方法:One is the method of manually assigning weights:

cT的词的权重可以这样赋予:对于主题词赋予最大的权重,对于强相关词赋予中间大小的权重,对于弱相关词赋予最小权重。The weight of the words of c T can be given as follows: give the largest weight to the subject words, give the middle weight to the strong related words, and give the smallest weight to the weak related words.

比如:主题词(如餐饮领域cT的”川菜”)赋予权重1,强相关词(如餐饮领域cT的”辣”)赋予权重0.8,弱相关词(如餐饮领域cT的”香”)赋予权重0.5For example: subject words (such as "Sichuan Cuisine" in the catering field c T ) are given a weight of 1, strong related words (such as "spicy" in the catering field c T ) are given a weight of 0.8, weakly related words (such as "fragrant" in the catering field c T ) gives a weight of 0.5

另一种是通过学习自动分配权重的方法:The other is by learning to assign weights automatically:

对每个领域cT收集一些有代表性的训练文本语料资料;Collect some representative training text corpora for each domain c T ;

对语料样本进行切词,生存领域cT的词库;Segment the words of the corpus samples, the lexicon of c T in the survival field;

计算领域cT中的词的权重,权重=TF×GIDF,其中TF为词在该领域cT所有语料中的词的总词频,GIDF为全局反向文档频率,GIDF=log(1+N/GDF),其中N为所有领域的所有文档的总数量,GDF为全局文档频率即为所有领域中包含该词的的所有文档的数量。Calculate the weight of the words in the field c T , weight=TF×GIDF, where TF is the total word frequency of words in all corpus of the field c T , GIDF is the global reverse document frequency, GIDF=log(1+N/ GDF), where N is the total number of all documents in all fields, and GDF is the global document frequency, which is the number of all documents containing the word in all fields.

设置各个水平的阈值,如T1,T2,...,Tn,T1>T2>...>TnSet thresholds for each level, such as T1, T2, ..., Tn, T1>T2>...>Tn

对领域cT词库中词根据其权重按上面阈值划分为多个档次的集合,Ti>总词频>Ti+1的为第个档次。For the field c, the words in the T lexicon are divided into sets of multiple grades according to their weights according to the above threshold, and the first grade is Ti>total word frequency>Ti+1.

对各个档次的词分别赋予一定的最终评分值,第一档赋予最高评分值,中间档赋予中间大小的评分值,第n档赋予最小评分值。A certain final score value is assigned to each grade of words, the first grade is assigned the highest score value, the middle grade is assigned a score value of an intermediate size, and the nth grade is assigned a minimum score value.

由词库中的词及其最终评分值组成领域cT向量。The domain c T vector is composed of words in the thesaurus and their final score values.

给搜索请求的关键字赋予一定的权重,组成一个Query的向量,Query(q1,q2,...qn’)。Assign a certain weight to the keywords of the search request to form a Query vector, Query(q1, q2,...qn').

(2)Query的关键字的权重可以这样赋予:(2) The weight of the keyword of Query can be given as follows:

方法1:全部关键字赋予权重1;Method 1: assign weight 1 to all keywords;

方法2:排在最前面的关键字赋予最大权重(比如赋予权重1),排在中间的关键字赋予中间大小的权重(比如赋予0.5<权重<1),排在最后的关键字赋予最小权重(比如赋予权重0.5)。Method 2: The top keywords are given the greatest weight (for example, weight 1), the middle keywords are given middle weights (for example, 0.5<weight<1), and the last keywords are given the smallest weight (For example, assign a weight of 0.5).

(3)计算领域向量cT(t1,t2,...,tn)与查询向量qT(q1,q2,...,qn’)之间的Cousine相似度:(3) Calculate the Cousine similarity between domain vector c T (t1, t2, ..., tn) and query vector q T (q1, q2, ..., qn'):

SimSim (( qq TT (( qq 11 ,, qq 22 ,, .. .. .. ,, qnqn ,, )) ,, cc TT (( tt 11 ,, tt 22 ,, .. .. .. ,, tntn )) ))

== (( qq 11 &times;&times; tt 11 ++ qq 22 &times;&times; tt 22 ++ .. .. .. .. .. .. ++ qnqn &times;&times; tntn )) // (( qq 11 22 ++ qq 22 22 ++ .. .. .. ++ qnqn 22 &times;&times;

tt 11 22 ++ tt 22 22 ++ .. .. .. ++ tntn 22 ))

(4)g1(cT,qT)=sim(qT,cT);(4) g 1 (c T , q T ) = sim(q T , c T );

(5)从搜索历史相关数据中收集查询q(t)的所有用户点击历史文档UT={uT},其中uT为查询qT对应的某个用户点击搜索结果文档的向量,计算uT与cT的cousine相似度:(5) Collect all user click history documents U T ={u T } for query q(t) from the search history related data, where u T is the vector of a user click search result document corresponding to query q T , and calculate u Cousine similarity between T and c T :

simsim (( cc TT (( tt 11 ,, tt 22 ,, .. .. .. ,, tntn )) ,, uu TT (( uu 11 ,, uu 22 ,, .. .. .. ,, unun )) ))

== (( uu 11 &times;&times; tt 11 ++ uu 22 &times;&times; tt 22 ++ .. .. .. .. .. .. ++ unun &times;&times; tntn )) // (( uu 11 22 ++ uu 22 22 ++ .. .. .. ++ unun 22 &times;&times;

tt 11 22 ++ tt 22 22 ++ .. .. .. ++ tntn 22

(6) g 2 ( c T , q T ) = &Sigma; u T sim ( c T , u T ) | U T | (6) g 2 ( c T , q T ) = &Sigma; u T sim ( c T , u T ) | u T |

上下文相关的特征函数fk的选取:Selection of context-dependent feature function f k :

(1)直接关联(1) Direct association

设置查询序列对(qt-1,qt)的标记序列对为(ct-1,ct),本实施例用在给定查询序列对(qt-1,qt)前提下,标记序列对(ct-1,ct)出现的次数来计算f1(ct-1,ct,q)Set the tag sequence pair of the query sequence pair (q t-1 , q t ) to (c t-1 , c t ), this embodiment is used on the premise of a given query sequence pair (q t-1 , q t ), Count the occurrences of the sequence pair (c t-1 , c t ) to compute f 1 (c t-1 , c t , q)

ff 11 (( cc tt -- 11 ,, cc tt ,, qq )) == Oo (( cc tt -- 11 ,, cc tt )) Oo (( qq tt -- 11 ,, qq tt ))

其中O(ct-1,ct)为用在给定查询序列对(qt-1,qt)前提下,标记序列对(ct-1,ct)出现的次数。Where O(c t-1 , c t ) is the number of occurrences of the marker sequence pair (c t-1 , c t ) under the premise of a given query sequence pair (q t-1 , q t ).

O(qt-1,qt)为查询序列对(qt-1,qt)出现的总次数。O(q t-1 , q t ) is the total number of occurrences of the query sequence pair (q t-1 , q t ).

(2)利用分类目录树间接关联(2) Indirect association using classification tree

假设标记序列对(ct-1,ct)处于分类目录树的第n层,(ct-1,ct)的祖先节点对的集合为

Figure B2009102208107D0000175
1≤i≤n-1,本实施例用在给定查询序列对(qt-1,qt)前提下(ct-1,ct)的祖先节点对
Figure B2009102208107D0000176
出现的次数来计算f2(ct-1,ct,q):Assuming that the tag sequence pair (c t-1 , c t ) is at the nth level of the classification tree, the set of ancestor node pairs of (c t-1 , c t ) is
Figure B2009102208107D0000175
1≤i≤n-1, this embodiment uses the ancestor node pair of (c t-1 , c t ) under the premise of a given query sequence pair (q t-1 , q t )
Figure B2009102208107D0000176
occurrences to calculate f 2 (c t-1 , c t , q):

ff 22 (( cc tt -- 11 ,, cc tt ,, qq )) == &Sigma;&Sigma; ii == 11 nno -- 11 Oo (( aa cc tt -- 11 (( ii )) ,, aa cc tt (( ii )) )) Oo (( qq tt -- 11 ,, qq tt ))

其中,

Figure B2009102208107D0000178
为在给定查询序列对(qt-1,qt)前提下(ct-1,ct)的祖先节点对
Figure B2009102208107D0000179
出现的次数,O(qt-1,qt)为查询序列对(qt-1,qt)出现的总次数。in,
Figure B2009102208107D0000178
is the ancestor node pair of (c t-1 , c t ) given the query sequence pair (q t-1 , q t )
Figure B2009102208107D0000179
The number of occurrences, O(q t-1 , q t ) is the total number of occurrences of the query sequence pair (q t-1 , q t ).

进一步地,图1中所述步骤102可以包括:计算所述关键词信息与成员引擎的元索引信息之间的第一最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第一阈值且成员引擎的元索引信息与长期兴趣模型的相似度大于第二阈值的基础上,成员引擎的元索引信息与即时兴趣模型的第二最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第三阈值且成员引擎的元索引信息与即时兴趣模型的相似度大于第四阈值的基础上,成员引擎的元索引信息与长期兴趣模型的第三最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第五阈值的基础上,成员引擎的元索引信息与长期兴趣模型和即时兴趣模型的加权相加的结果向量的第四最大相似度;根据第一最大相似度、第二最大相似度、第三最大相似度和第四最大相似度计算成员引擎的相似度评分值。Further, the step 102 in FIG. 1 may include: calculating the first maximum similarity between the keyword information and the meta index information of the member engine; calculating the similarity between the meta index information of the member engine and the keyword information If it is greater than the first threshold and the similarity between the meta index information of the member engine and the long-term interest model is greater than the second threshold, the second maximum similarity between the meta index information of the member engine and the instant interest model; calculated in the meta index of the member engine On the basis that the similarity between information and keyword information is greater than the third threshold and the similarity between the meta index information of the member engine and the instant interest model is greater than the fourth threshold, the third maximum similarity between the meta index information of the member engine and the long-term interest model ; Calculate the fourth maximum similarity between the meta index information of the member engine and the result vector of the weighted addition of the long-term interest model and the instant interest model on the basis that the similarity between the meta index information of the member engine and the keyword information is greater than the fifth threshold Degree; Calculate the similarity score value of the member engine according to the first maximum similarity, the second maximum similarity, the third maximum similarity and the fourth maximum similarity.

需要说明的是,上述第一阈值、第二阈值、第三阈值、第四阈值以及第五阈值的取值范围为0~1,本领域技术人员可以根据需要设定这些阈值的取值,或者采用经验值。It should be noted that the value ranges of the above-mentioned first threshold, second threshold, third threshold, fourth threshold and fifth threshold are 0 to 1, and those skilled in the art can set the values of these thresholds according to needs, or Use experience.

具体来说,获取所述选择结果信息的步骤可以如下:Specifically, the steps of obtaining the selection result information may be as follows:

假设成员引擎对应的数据库为D,用户的当前查询向量为Q(q1,...,qk-1,qk),用户的即时兴趣模型向量为IIM(u1,....un-1,un)(Instant interest model)、用户的长期兴趣模型向量为LIM(r1,....rn-1,rn)(Long-term interest model).Suppose the database corresponding to the membership engine is D, the user's current query vector is Q(q 1 ,...,q k-1 ,q k ), and the user's instant interest model vector is IIM(u 1 ,...u n-1 ,u n )(Instant interest model), the user's long-term interest model vector is LIM(r 1 ,...r n-1 ,r n )(Long-term interest model).

(a)计算Q与D的最大相似度,即第一最大相似度sim1(a) Calculate the maximum similarity between Q and D, that is, the first maximum similarity sim 1 .

Figure B2009102208107D0000181
Figure B2009102208107D0000182
的值,其中|Q|为搜索请求向量Q的模,|R|为用户的兴趣模型R的模;该值作为该第一最大相似度。
Figure B2009102208107D0000181
Figure B2009102208107D0000182
, where |Q| is the modulus of the search request vector Q, and |R| is the modulus of the user's interest model R; this value is used as the first maximum similarity.

(b)计算在D与Q以及D与LIM分别取得较好匹配的基础上,D与IIM的第二最大相似度sim2 (b) Calculate the second maximum similarity sim 2 between D and IIM on the basis of better matching between D and Q and D and LIM respectively

simsim 22 == MaxMax 11 &le;&le; ii &le;&le; nno (( ifif (( simsim (( VV (( mnvmnv ii ,, anvanv jj (( jj &NotEqual;&NotEqual; ii ,, 11 &le;&le; jj &le;&le; nno )) )) ,, QQ ,, )) >> TT 11 andand

simsim (( VV (( mnvmnv ii ,, anvanv jj (( jj &NotEqual;&NotEqual; ii ,, 11 &le;&le; jj &le;&le; nno )) )) ,, LIMLIM (( rr 11 ,, .. .. .. .. rr nno -- 11 ,, rr nno )) )) >> TT 22 )) thenthen

(( (( uu ii ** mnvmnv ii ++ &Sigma;&Sigma; jj == 11 ,, jj &NotEqual;&NotEqual; ii nno uu jj ** anvanv jj )) // || RR || ++ &Sigma;&Sigma; ii == 11 kk qq ii ** gidfgidf ii ** anwanw ii // || QQ || ))

其中,Q’的计算方法为:如果术语ti属于用户的兴趣模型的某个维度的范围,将qi的值映射成用户的兴趣模型的该维度的权重,然后将相同维度的权重相加得到qi’,再作归一化处理;V为由mnvi和anvj(j≠i,1≤j≤n)组成的向量;sim(V(mnvi,anvj(j≠i,1≤j≤n)),Q’)为向量V和向量Q’的cousine相似度;sim(V(mnvi,anvj(j≠i,1≤j≤n)),LIM(r1,....rn-1,rn))为向量V和向量LIM的cousine相似度;T1,T2为阈值,且0<T1,T2≤1;i,k,j,n为自然数。Among them, the calculation method of Q' is: if the term ti belongs to the scope of a certain dimension of the user's interest model, the value of qi is mapped to the weight of this dimension of the user's interest model, and then the weights of the same dimension are added to get qi ', then normalized; V is a vector composed of mnvi and anvj (j≠i, 1≤j≤n); sim(V(mnvi, anvj(j≠i, 1≤j≤n)), Q') is the Cousine similarity between vector V and vector Q'; sim(V(mnv i , anv j (j≠i, 1≤j≤n)), LIM(r 1 ,...r n-1 , r n )) is the cousin similarity between vector V and vector LIM; T 1 , T 2 are thresholds, and 0<T 1 , T 2 ≤1; i, k, j, n are natural numbers.

(c)计算在D与Q以及D与IIM分别取得较好匹配的基础上,D与LIM的第三最大相似度sim3(c) Calculate the third maximum similarity sim 3 between D and LIM on the basis of good matching between D and Q and D and IIM respectively.

simsim 33 == MaxMax 11 &le;&le; ii &le;&le; nno (( ifif (( simsim (( VV (( mnvmnv ii ,, anvanv jj (( jj &NotEqual;&NotEqual; ii ,, 11 &le;&le; jj &le;&le; nno )) )) ,, QQ ,, )) >> TT 11 andand

simsim (( VV (( mnvmnv ii ,, anvanv jj (( jj &NotEqual;&NotEqual; ii ,, 11 &le;&le; jj &le;&le; nno )) )) ,, IIMIIM (( uu 11 ,, .. .. .. .. uu nno -- 11 ,, uu nno )) )) >> TT 22 )) thenthen

(( (( rr ii ** mnvmnv ii ++ &Sigma;&Sigma; jj == 11 ,, jj &NotEqual;&NotEqual; ii nno rr jj ** anvanv jj )) // || RR || ++ &Sigma;&Sigma; ii == 11 kk qq ii ** gidfgidf ii ** anwanw ii // || QQ || ))

其中,Q’的计算方法为:如果术语ti属于用户的兴趣模型的某个维度的范围,将qi的值映射成用户的兴趣模型的该维度的权重,然后将相同维度的权重相加得到qi’,再作归一化处理;V为由mnvi和anvj(j≠i,1≤j≤n)组成的向量;sim(V(mnvi,anvj(j≠i,1≤j≤n)),Q’)为向量V和向量Q’的cousine相似度;sim(V(mnvi,anvj(j≠i,1≤j≤n)),IIM(u1,....un-1,un)为向量V和向量IIM的cousine相似度;T1,T2为阈值,且0<T1,T2≤1;i,k,j,n为自然数。Among them, the calculation method of Q' is: if the term ti belongs to the scope of a certain dimension of the user's interest model, the value of qi is mapped to the weight of this dimension of the user's interest model, and then the weights of the same dimension are added to get qi ', then normalized; V is a vector composed of mnvi and anvj (j≠i, 1≤j≤n); sim(V(mnvi, anvj(j≠i, 1≤j≤n)), Q') is the Cousine similarity between vector V and vector Q'; sim(V(mnv i , anv j (j≠i, 1≤j≤n)), IIM(u 1 ,...u n-1 , u n ) is the cousin similarity between vector V and vector IIM; T 1 , T 2 are thresholds, and 0<T 1 , T 2 ≤1; i, k, j, n are natural numbers.

(d)计算在D与Q取得较好匹配的基础上,D与IIM和LIM加权相加所的结果向量之间的第四最大相似度sim4(d) Calculate the fourth maximum similarity sim 4 between D and the result vector of the weighted addition of IIM and LIM on the basis of better matching between D and Q.

设IM(p1,...,pn-1,pn)=r′1LIM(r1,...,rn)+r′2IIM(u1,...,un-1,un),r′1+r′2=1,Let IM(p 1 ,...,p n-1 ,p n )=r' 1 LIM(r 1 ,...,r n )+r' 2 IIM(u 1 ,...,u n- 1 , u n ), r′ 1 +r′ 2 =1,

simsim 44 == MaxMax 11 &le;&le; ii &le;&le; nno (( ifif (( simsim (( VV (( mnvmnv ii ,, anvanv jj (( jj &NotEqual;&NotEqual; ii ,, 11 &le;&le; jj &le;&le; nno )) )) ,, QQ ,, )) >> TT 11 )) thenthen

(( (( pp ii ** mnvmnv ii ++ &Sigma;&Sigma; jj == 11 ,, jj &NotEqual;&NotEqual; ii nno pp jj ** anvanv jj )) // || RR || ++ &Sigma;&Sigma; ii == 11 kk qq ii ** gidfgidf ii ** anwanw ii // || QQ || ))

(e)根据sim1~sim4计算成员引擎的相似度评分值Msim:(e) Calculate the similarity score value Msim of the member engine according to sim1~sim4:

方法1:取sim1~sim4的最大值.Method 1: Take the maximum value of sim1~sim4.

Msim=Max{sim1,sim2,sim3,sim4}Msim=Max{sim1, sim2, sim3, sim4}

方法2:sim1~sim4加权相加Method 2: Weighted addition of sim1 to sim4

Msim=r1×sim1+r2×sim2+r3×sim3+r4×sim4,其中ri+r2+r3+r4=1.Msim=r1×sim1+r2×sim2+r3×sim3+r4×sim4, where ri+r2+r3+r4=1.

方法3:sim1~sim4相乘Method 3: multiply sim1~sim4

Msim=sim1×sim2×sim3×sim4Msim=sim1×sim2×sim3×sim4

除了上面方法计算成员引擎相关度评分值外,还可以进一步考虑其他相关因素,如成员引擎的性能因素和成员引擎的价格因素等,最终的成员引擎综合相关度评分值为上述基于元索引的相似度评分值(Msim)与性能因素评分值和价格因素评分值的加权相加:In addition to the calculation of the member engine relevance score value by the above method, other related factors can be further considered, such as the performance factor of the member engine and the price factor of the member engine, etc. The weighted addition of degree score value (Msim) and performance factor score value and price factor score value:

Integrated_Sim=r1×Msim+r2×性能因素评分值+r3×价格因素评分值,其中r1+r2+r3=1Integrated_Sim=r1×Msim+r2×performance factor score value+r3×price factor score value, where r1+r2+r3=1

对于每个成员引擎,均可以采用上述步骤(a)~(e)进行操作,从而可以获取每个成员引擎的评分最大值。For each member engine, the above steps (a) to (e) can be used to perform operations, so that the maximum score of each member engine can be obtained.

在选择使用哪个成员引擎搜索关键词信息时,可以根据需要从这些评分最大值的集合中选择前一个或者前几个评分最大值,即选择一个或多个成员引擎,从而使得搜索服务器可以将搜索请求分发给选择出的成员引擎进行搜索操作。在具体实现过程中,可以获取选择出的成员引擎的ID信息作为选择结果信息,从而向与ID信息对应的成员引擎发送搜索请求。When choosing which member engine to use to search for keyword information, you can select the previous one or several previous maximum score values from the set of these maximum score values according to your needs, that is, select one or more member engines, so that the search server can search Requests are distributed to selected member engines for search operations. In a specific implementation process, the ID information of the selected member engine may be acquired as selection result information, so as to send a search request to the member engine corresponding to the ID information.

下面以一个具体实施例对本发明移动搜索方法上述实施例的技术方案进行详细说明。The technical solutions of the above embodiments of the mobile search method of the present invention will be described in detail below with a specific embodiment.

图2为本发明移动搜索方法实施例二的信令流程图,如图2所示,本实施例的方法可以包括:FIG. 2 is a signaling flow chart of Embodiment 2 of the mobile search method of the present invention. As shown in FIG. 2, the method of this embodiment may include:

步骤201、搜索服务器接收各个成员引擎上报的元索引信息。Step 201, the search server receives the meta index information reported by each member engine.

需要说明的是,该步骤为可选的,且其执行顺序不限于此。It should be noted that this step is optional, and its execution sequence is not limited thereto.

该元索引信息可以包括下述信息之一或者其任意组合:The meta index information may include one or any combination of the following information:

(1)术语最大归一化权重向量mnw=(mnw1,mnw2,...,mnwi,....,mnwp),其中mnwi为术语ti相对于该成员引擎对应的数据库或者子数据库中的所有文档的最大归一化权重。则其中,mnwi可以以下面的方式计算得到:首先计算数据库/子数据库中的每个文档相对于术语ti的归一化权重,归一化权重的取值可以为文档中术语ti出现的次数(词频)除以文档的长度,文档中术语ti的归一化权重=TFi/|d|,其中文档长度(tf1~tfn为文档的所有术语的词频),TFi为术语ti的词频。再从数据库/子数据库中所有文档相对术语ti的归一化权重中取最大值,得到数据库/子数据库术语t1的最大归一化权重。(1) Term maximum normalized weight vector mnw=(mnw1, mnw2, ..., mnwi, ..., mnwp), where mnwi is term ti relative to all the members in the database or sub-database corresponding to the member engine The maximum normalized weight for a document. Among them, mnwi can be calculated in the following manner: first calculate the normalized weight of each document in the database/subdatabase relative to the term ti, and the value of the normalized weight can be the number of times ti appears in the document ( term frequency) divided by the length of the document, the normalized weight of the term ti in the document = TFi/|d|, where the document length (tf1~tfn are the word frequencies of all terms in the document), and TFi is the word frequency of the term ti. Then take the maximum value from the normalized weights of all documents in the database/subdatabase relative to the term ti to obtain the maximum normalized weight of the database/subdatabase term t1.

(2)术语平均归一化权重向量anw=(anw1,anw2,.....,anwi......,anwp),其中anwi为术语ti相对于该成员引擎对应的数据库或者子数据库中的所有文档的平均归一化权重。则其中,anwi可以以下面的方式计算得到:首先计算数据库/子数据库中的每个文档相对于术语ti的归一化权重,归一化权重的取值可以为文档中术语ti出现的次数(词频)除以文档的长度,文档中术语ti的归一化权重=TFi/|d|,其中文档长度

Figure B2009102208107D0000212
(tf1~tfn为文档的所有术语的词频),TFi为术语ti的词频。再从数据库/子数据库中所有文档相对术语ti的归一化权重中取平均值,得到数据库/子数据库术语t1的平均归一化权重。(2) Term average normalized weight vector anw=(anw1, anw2, ..., anwi..., anwp), where anwi is the database or sub-database corresponding to the member engine for the term ti The average normalized weight of all documents in . Wherein, anwi can be calculated in the following manner: first calculate the normalized weight of each document in the database/subdatabase relative to the term ti, and the value of the normalized weight can be the number of occurrences of the term ti in the document ( term frequency) divided by the length of the document, the normalized weight of the term ti in the document = TFi/|d|, where the document length
Figure B2009102208107D0000212
(tf1~tfn are the word frequencies of all terms in the document), and TFi is the word frequency of the term ti. Then average the normalized weights of all documents in the database/subdatabase relative to term ti to obtain the average normalized weight of term t1 in the database/subdatabase.

(3)数据库或者子数据库中的文档的兴趣模型最大归一化权重向量mnv=(mnv1,mnv2,......,mnvi,......,mnvn),其中mnvi为该文档的兴趣模型的第i个维度相对于该成员引擎对应的数据库或者子数据库中的所有文档的最大归一化权重。其中,mnvi可以通过下面的方式计算得到:(3) The maximum normalized weight vector mnv=(mnv1, mnv2, ..., mnvi, ..., mnvn) of the interest model of the document in the database or sub-database, where mnvi is the document The i-th dimension of the interest model of is relative to the maximum normalized weight of all documents in the database or sub-database corresponding to the membership engine. Among them, mnvi can be calculated in the following way:

方法1:首先计算数据库的每个文档相对于兴趣模型第i个维度的归一化权重,归一化权重的取值为文档中属于兴趣模型第i个维度范围(如:体育)的所有词的词频之和再除以文档的长度;再从所有文档相对于兴趣模型第i个维度的归一化权重中取最大值,就得到兴趣模型的第i个维度相对于数据库D中的所有文档的最大归一化权重mnvi。Method 1: First calculate the normalized weight of each document in the database relative to the i-th dimension of the interest model, and the value of the normalized weight is all words in the document that belong to the i-th dimension of the interest model (eg: sports) Divide the sum of the word frequency of the interest model by the length of the document; then take the maximum value from the normalized weights of all documents relative to the i-th dimension of the interest model, and then get the i-th dimension of the interest model relative to all documents in the database D The maximum normalized weight mnvi of .

方法2:首先计算数据库的每个文档相对于兴趣模型第i个维度的归一化权重,将文档进行自动分类(分类的方法可以采用常用的朴素贝叶斯、K最近邻分类算法、支持向量机、向量空间模型等算法),将文档属于第i个维度对应类型的归一化评分值作为每个文档相对于兴趣模型第i个维度的归一化权重的取值,再从所有文档相对于兴趣模型第i个维度的归一化权重中取最大值,就得到兴趣模型的第i个维度相对于数据库D中的所有文档的最大归一化权重mnvi。Method 2: First calculate the normalized weight of each document in the database relative to the i-th dimension of the interest model, and automatically classify the documents (classification methods can use commonly used naive Bayesian, K-nearest neighbor classification algorithms, support vector machine, vector space model and other algorithms), the normalized score value of the document belonging to the i-th dimension corresponding to the type is used as the value of the normalized weight of each document relative to the i-th dimension of the interest model, and then from all documents relative to Taking the maximum value from the normalized weight of the i-th dimension of the interest model, the maximum normalized weight mnvi of the i-th dimension of the interest model relative to all documents in the database D is obtained.

(4)数据库或者子数据库中的文档的兴趣模型平均归一化权重向量anv=(anv1,anv2,......,anvi,......,anvn),其中anvi为该文档的兴趣模型的第i个维度相对于该成员引擎对应的数据库中的所有文档的平均归一化权重。其中,anvi可以通过下面的方式计算得到:(4) The average normalized weight vector anv=(anv1, anv2, ..., anvi, ..., anvn) of the interest model of the document in the database or sub-database, where anvi is the document The i-th dimension of the interest model is relative to the average normalized weight of all documents in the database corresponding to the membership engine. Among them, anvi can be calculated in the following way:

方法1:首先计算数据库的每个文档相对于兴趣模型第i个维度的归一化权重,归一化权重的取值为文档中属于兴趣模型第i个维度范围(如:体育)的所有词的词频之和再除以文档的长度;再从所有文档相对于兴趣模型第i个维度的归一化权重中取平均值,就得到兴趣模型的第i个维度相对于数据库D中的所有文档的平均归一化权重anvi。Method 1: First calculate the normalized weight of each document in the database relative to the i-th dimension of the interest model, and the value of the normalized weight is all words in the document that belong to the i-th dimension of the interest model (eg: sports) Divide the sum of word frequencies by the length of the document; then take the average from the normalized weights of all documents relative to the i-th dimension of the interest model, and then get the i-th dimension of the interest model relative to all documents in the database D The average normalized weight of anvi.

方法2:首先计算数据库的每个文档相对于兴趣模型第i个维度的归一化权重,将文档进行自动分类(分类的方法可以采用常用的朴素贝叶斯、K最近邻分类算法、支持向量机、向量空间模型等算法),将文档属于第i个维度对应类型的归一化评分值作为每个文档相对于兴趣模型第i个维度的归一化权重的取值,;再从所有文档相对于兴趣模型第i个维度的归一化权重中取平均值,就得到兴趣模型的第i个维度相对于数据库D中的所有文档的平均归一化权重anviMethod 2: First calculate the normalized weight of each document in the database relative to the i-th dimension of the interest model, and automatically classify the documents (classification methods can use commonly used naive Bayesian, K-nearest neighbor classification algorithms, support vector machine, vector space model, etc.), the normalized score value of the document belonging to the i-th dimension corresponding to the type is used as the value of the normalized weight of each document relative to the i-th dimension of the interest model; and then from all documents Taking the average of the normalized weights of the i-th dimension of the interest model, the average normalized weight anvi of the i-th dimension of the interest model relative to all documents in the database D is obtained

(5)术语ti相对于该数据库的全局反向文档频率gidfi,其中gidfi=log(n/(gdfi+1)),其中gdfi为所有成员引擎对应数据库或者子数据库中包含术语ti的文档的数量的总和,n为所有成员引擎所包含的所有文档数量的总和;(5) The global inverse document frequency gidfi of the term ti relative to the database, where gidfi=log(n/(gdfi+1)), where gdfi is the number of documents containing the term ti in the corresponding database or sub-database of all member engines The sum of , n is the sum of the number of all documents contained in all member engines;

(6)文档的兴趣模型第i个维度对应的全局反向文档频率IM_gidfi,其中IM_gidfi=log(n/(IM_gdfi+1)),IM_gdfi为所有成员引擎对应的数据库或子数据库中包含属于文档的兴趣模型的第i个维度的术语的文档个数的总和,n为所有成员引擎所包含的所有文档数量的总和。(6) The global inverse document frequency IM_gidfi corresponding to the i-th dimension of the document interest model, where IM_gidfi=log(n/(IM_gdfi+1)), IM_gdfi is the database or sub-database corresponding to all member engines that contains documents The sum of the number of documents of the term in the i-th dimension of the interest model, n is the sum of the number of all documents contained in all member engines.

步骤202、搜索客户端将搜索请求发给搜索应用服务器。Step 202, the search client sends a search request to the search application server.

步骤203、搜索应用服务器从用户数据库中提取用户的长期兴趣模型和即时兴趣模型。Step 203, the search application server extracts the user's long-term interest model and immediate interest model from the user database.

例如,搜索应用服务器可以从用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前查询q(t)处于同一搜索上下文会话(Search Context Session)的查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型,所述搜索上下文会话为当前查询q(t)发生的前一段预设的时间,如半个小时,包括q(t)当前发生的时间。For example, the search application server can extract the user's long-term interest model from the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract the user's long-term interest model from the current query q( t) Extract the user's instant interest model from the relevant data of the query sequence q(1), ..., q(t-1), q(t) in the same search context session (Search Context Session), the search context A session is a preset period of time before the occurrence of the current query q(t), such as half an hour, including the current occurrence time of q(t).

对于用户的静态profile所对应的长期兴趣模型W1来说,W1=(p1,p2,p3,......,pn),其中pi为静态profile中类型属于第i个兴趣维度的所有词的词频之和;或者对该静态profile对应的文档进行分类(分类的算法可以用朴素贝叶斯、K最近邻分类算法、支持向量机、向量空间模型等算法),tj等于该静态profile对应的文档属于第j个兴趣维度所对应的类型的评分值。For the long-term interest model W1 corresponding to the user's static profile, W1=(p1, p2, p3, ..., pn), where pi is all the words in the static profile that belong to the i-th interest dimension or classify the documents corresponding to the static profile (the classification algorithm can use Naive Bayesian, K nearest neighbor classification algorithm, support vector machine, vector space model and other algorithms), tj is equal to the static profile corresponding to The document belongs to the rating value of the type corresponding to the jth interest dimension.

对于用户的搜索点击历史所对应的长期兴趣模型W2来说,W2=d1+d2+d3+......dm,其中di为用户某个点击文档所对应的兴趣模型向量,di=(t1,t2,t3,.......,tn),当用户最新点击了这个文档,tj等于文档中类型属于第j个兴趣维度的所有词的词频之和;或者对该文档进行分类(分类的算法可以用朴素贝叶斯、K最近邻分类算法、支持向量机、向量空间模型等算法),tj等于该文档属于第j个兴趣维度所对应的类型的评分值。如果用户对某个点击过的文档进行评价,如果评价好,di向量乘以一个正的常数c表示文档的重要性增加di=c×di=(c×ti,c×t2,c×t3,......,c×tn),如果评价不好,di向量乘以一个正的常数c的倒数表示文档的重要性减小di=1/c×di=(1/c×ti,1/c×t2,1/c×t3,......,1/c×tn);过了一段时间,tj的值又自动减少一定的百分比,表示随着时间的推移其重要性减弱,直到过了较长的时间tj的值减为零为止,这时将di从历史记录中删除。For the long-term interest model W2 corresponding to the user's search click history, W2=d1+d2+d3+...dm, where di is the interest model vector corresponding to a document clicked by the user, di=(t1 , t2, t3,......, tn), when the user clicks on this document recently, tj is equal to the sum of the word frequencies of all words in the document whose type belongs to the jth dimension of interest; or classify the document ( The classification algorithm can use naive Bayesian, K nearest neighbor classification algorithm, support vector machine, vector space model and other algorithms), and tj is equal to the score value of the type corresponding to the jth dimension of interest that the document belongs to. If the user evaluates a document that has been clicked, if the evaluation is good, the di vector is multiplied by a positive constant c to indicate that the importance of the document increases di=c×di=(c×ti, c×t2, c×t3, ......, c×tn), if the evaluation is not good, the reciprocal of the di vector multiplied by a positive constant c indicates that the importance of the document is reduced di=1/c×di=(1/c×ti, 1/c×t2, 1/c×t3, ..., 1/c×tn); after a period of time, the value of tj will automatically decrease by a certain percentage, indicating its importance over time Weaken until the value of tj decreases to zero after a long period of time, then delete di from the history.

对于综合的长期兴趣模型来说,可以将W1和W2分别归一化后相加,即:兴趣模型向量W=W1+W2,或者加权相加,如兴趣模型向量W=W1×30%+W2×70%,然后再对W进行归一化处理。For a comprehensive long-term interest model, W1 and W2 can be normalized and added separately, that is: interest model vector W=W1+W2, or weighted addition, such as interest model vector W=W1×30%+W2 ×70%, and then normalize W.

步骤204、搜索应用服务器将搜索请求发送给搜索服务器。Step 204, the search application server sends the search request to the search server.

该搜索请求中携带即时兴趣模型和长期兴趣模型。The search request carries an immediate interest model and a long-term interest model.

步骤205、搜索服务器根据搜索请求选择成员引擎。Step 205, the search server selects a member engine according to the search request.

搜索服务器可以根据即时兴趣模型和长期兴趣模型以及步骤201获取的元索引信息计算成员引擎对应的数据库的相似度评分值,选择相似度评分值高的成员引擎。选择成员引擎,获取选择结果信息的方法可以采用上述描述的过程,不再赘述。The search server may calculate the similarity score value of the database corresponding to the member engine according to the immediate interest model, the long-term interest model and the meta-index information obtained in step 201, and select a member engine with a high similarity score. The method for selecting a member engine and obtaining the selection result information may adopt the process described above, which will not be repeated here.

步骤206、搜索服务器将搜索请求分发给选中的成员引擎。Step 206, the search server distributes the search request to the selected member engines.

举例来说,搜索服务器获取的选择结果信息可以为选中的成员引擎的ID信息,因此该搜索服务器可以向与ID信息对应的成员引擎分发搜索请求,从而使得一个或多个成员引擎可以对该关键词信息进行搜索,从而获取搜索结果信息。For example, the selection result information obtained by the search server can be the ID information of the selected member engine, so the search server can distribute search requests to the member engines corresponding to the ID information, so that one or more member engines can use the key Word information is searched to obtain search result information.

本实施例中,搜索服务器在请求成员引擎搜索关键词信息之前,可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎进行选择,从而能够选择获取与所需搜索的关键词信息以及用户的即时兴趣模型和长期兴趣模型匹配较好的成员引擎对该关键词信息进行搜索,从而能够获取精度较高的搜索结果信息,进一步满足用户的搜索需求。In this embodiment, before the search server requests the member engine to search for keyword information, it can select the member engine according to the user's immediate interest model and long-term interest model, so as to be able to select and obtain keyword information related to the desired search and the user's real-time interest model. The member engine whose interest model and long-term interest model match well searches for the keyword information, so as to obtain search result information with high precision, and further meet the user's search needs.

图3为本发明移动搜索方法实施例三的流程图,如图3所示,本实施例的方法可以包括:FIG. 3 is a flow chart of Embodiment 3 of the mobile search method of the present invention. As shown in FIG. 3 , the method of this embodiment may include:

步骤301、向一个或多个成员引擎发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。Step 301: Send a search request to one or more member engines, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server.

搜索服务器可以向一个或多个成员引擎发送搜索请求,以指示接收到该搜索请求的成员引擎可以根据搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型进行搜索,获取具有个性化评分信息的搜索结果信息。The search server can send a search request to one or more member engines to indicate that the member engine that receives the search request can use the search keyword information carried in the search request and the immediate interest model and long-term interest model acquired by the search application server Do a search and get search result information with personalized scoring information.

举例来说,用户的兴趣模型可以用n个维度来表示如:新闻、体育、娱乐、财经、科技、房产、游戏、女性、论坛、天气、商品、家电、音乐、读书、博客、手机、军事、教育、旅游、彩信、彩铃、餐饮、民航、工业、农业、电脑、地理等。用户对每个维度的兴趣的评分值所组成的一个向量W(r1,r2,r3,......,rn)则为用户的兴趣模型。For example, the user's interest model can be represented by n dimensions such as: news, sports, entertainment, finance, technology, real estate, games, women, forums, weather, commodities, home appliances, music, reading, blogs, mobile phones, military , education, tourism, MMS, CRBT, catering, civil aviation, industry, agriculture, computer, geography, etc. A vector W(r1, r2, r3, . . . , rn) composed of rating values of the user's interest in each dimension is the user's interest model.

如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由用户的所有搜索历史数据和用户的静态档案profile计算得到,则兴趣模型W(r1,r2,r3,...rn)为用户的长期兴趣模型。如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到的,那么W(r1,r2,r3,...rn)为用户的即时兴趣模型。If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is calculated from all the user's search history data and the user's static profile profile, then the interest model W(r1, r2 , r3,...rn) is the user's long-term interest model. If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is determined by the query sequence q(1) in the same search context session as the current query q(t), ... , q(t-1), q(t) related data is calculated, then W(r1, r2, r3,...rn) is the user's instant interest model.

步骤302、接收所述一个或多个成员引擎根据所述关键词信息、所述即时兴趣模型和长期兴趣模型获取的搜索结果信息以及与所述搜索结果信息对应的评分信息。Step 302: Receive the search result information obtained by the one or more member engines according to the keyword information, the immediate interest model and the long-term interest model, and score information corresponding to the search result information.

成员引擎在接收该搜索请求后,可以对所需搜索的关键词信息进行搜索,获取搜索结果信息。然后,成员引擎可以根据即时兴趣模型和长期兴趣模型对搜索结果信息进行个性化的评分处理,从而获取与各搜索结果信息对应的具有个性化的评分信息,该评分信息可以表示各搜索结果信息与即时兴趣模型和长期兴趣模型之间的匹配程度。成员引擎可以将该搜索结果信息以及相应的评分信息发送给搜索服务器。After receiving the search request, the member engine can search for the keyword information to be searched to obtain search result information. Then, the member engine can perform personalized scoring processing on the search result information according to the instant interest model and the long-term interest model, thereby obtaining personalized scoring information corresponding to each search result information, and the scoring information can represent the relationship between each search result information and The degree of match between the immediate interest model and the long-term interest model. The member engine can send the search result information and corresponding scoring information to the search server.

步骤303、根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息发送给所述搜索应用服务器。Step 303: Perform re-scoring and sorting of the search result information according to the scoring information and relevant factor information, obtain the re-scoring and sorting search result information, and send the re-scoring and sorting search result information to the searcher application server.

搜索服务器在接收到各成员引擎反馈的搜索结果信息以及相应的评分信息后,可以对搜索结果信息进行重新评分处理,即对该搜索结果信息进行进一步的筛选处理,从而可以获取更为个性化地,符合需求的搜索结果信息。在重新评分处理过程中,搜索服务器可以将成员引擎反馈的与各搜索结果信息对应的评分信息以及其他相关因素信息结合起来对搜索结果信息进行综合评分。该相关因素信息可以包括该成员引擎的价格、级别、搜索速度、好评率等等信息,本领域技术人员可以根据需要将各种可能影响排序的信息结合到重新评分处理的过程中。After the search server receives the search result information and the corresponding scoring information fed back by each member engine, it can re-score the search result information, that is, further filter the search result information, so as to obtain more personalized results. , the search result information that meets the requirements. During the re-scoring process, the search server may combine the scoring information corresponding to each search result information fed back by the member engine and other relevant factor information to give a comprehensive score to the search result information. The relevant factor information may include information such as the member engine's price, level, search speed, favorable rate, etc. Those skilled in the art may combine various information that may affect the ranking into the re-rating process as required.

搜索服务器在完成重新评分处理后,即可将该重新评分处理后的搜索结果信息反馈给搜索应用服务器,以便搜索应用服务器可以将个性化的,满足用户需求的以及匹配精度高的搜索结果信息提供给搜索客户端。After the search server completes the re-scoring process, it can feed back the re-scored search result information to the search application server, so that the search application server can provide personalized search result information that meets user needs and has high matching accuracy. to the search client.

本实施例中,搜索服务器可以将搜索应用服务器提取的即时兴趣模型和长期兴趣模型发送给成员引擎,从而使得成员引擎在获取搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化评分处理,从而获取各搜索结果信息相应的评分信息。当搜索服务器接收到成员引擎反馈的搜索结果信息以及相应的评分信息以后,还可以结合其它相关因素对搜索结果信息进行重新评分排序,从而可以获取个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can send the immediate interest model and long-term interest model extracted by the search application server to the member engine, so that after the member engine obtains the search result information, it can use the user's immediate interest model and long-term interest model to the member engine. The search result information is subjected to personalized scoring processing, so as to obtain the corresponding scoring information of each search result information. After the search server receives the search result information fed back by the member engine and the corresponding scoring information, it can also re-score and sort the search result information in combination with other relevant factors, so as to obtain personalized, satisfying user needs and high matching accuracy. Search result information.

进一步地,本发明移动搜索方法另一个实施例还可以包括:接收所述搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。所述即时兴趣模型可以为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。所述长期兴趣模型为N个维度的评分值所组成的长期兴趣模型向量,各个维度的评分值由用户搜索历史数据和静态档案profile计算得到。Further, another embodiment of the mobile search method of the present invention may further include: receiving a search request sent by the search application server, the search request carrying keyword information to be searched and the instant interest model and Long-term interest model. The instant interest model can be an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are composed of the query sequence q(1) in the same search context session as the current query q(t),..., q(t-1), obtained by calculating the relevant data of q(t), the search context session is a period of preset time before the occurrence of the current query q(t) including the occurrence time of the current query q(t). The long-term interest model is a long-term interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are calculated from user search history data and static profile.

具体来说,搜索应用服务器可以根据搜索客户端发送的搜索请求消息,从用户数据库中获取用户的即时兴趣模型和长期兴趣模型。例如,从用户数据库,如用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前所需搜索的关键字信息处于同一查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型。在提取即时兴趣模型和长期兴趣模型后,搜索应用服务器即可向搜索服务器发送搜索请求,从而使得搜索服务器可以根据该搜索请求中携带的用户的即时兴趣模型和长期兴趣模型对所需搜索的关键字信息进行搜索。Specifically, the search application server can obtain the user's immediate interest model and long-term interest model from the user database according to the search request message sent by the search client. For example, extract the user's long-term interest model from the user database, such as the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract the user's long-term interest model from the current desired search The keyword information is in the same query sequence q(1), ..., q(t-1), q(t) related data to extract the user's instant interest model. After extracting the immediate interest model and the long-term interest model, the search application server can send a search request to the search server, so that the search server can determine the key information of the desired search according to the user's immediate interest model and long-term interest model carried in the search request. word information to search.

再进一步地,步骤303所述的相关因素信息可以包括:成员引擎级别信息和/或价格信息,相应地,步骤303中所述根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序可以包括根据所述评分信息、成员引擎级别信息和/或价格信息,计算所述搜索结果的综合评分值,并根据所述综合评分值对所述搜索结果信息进行排序处理。举例来说,搜索结果信息的综合评分值=r1×成员引擎返回评分值+r2×成员引擎级别相关的评分值+r3×价格因素相关的评分值,其中r1+r2+r3=1。Still further, the relevant factor information in step 303 may include: member engine level information and/or price information, and correspondingly, in step 303, the search result information is recreated according to the score information and relevant factor information Sorting by score may include calculating a comprehensive score value of the search result according to the score information, member engine level information and/or price information, and sorting the search result information according to the comprehensive score value. For example, the comprehensive score value of search result information=r1*return score value of member engine+r2*score value related to member engine level+r3*score value related to price factor, where r1+r2+r3=1.

需要说明的是,上述实施例也可以采用实施例一和实施例二所述的方法先行对所需使用的成员引擎进行选择处理,从而使得搜索服务器只向选中的成员引擎分发搜索请求,进一步提高搜索精度。It should be noted that, in the above-mentioned embodiment, the methods described in Embodiment 1 and Embodiment 2 can also be used to select the member engines to be used in advance, so that the search server only distributes search requests to the selected member engines, further improving Search precision.

本实施例中,搜索服务器可以将搜索应用服务器提取的即时兴趣模型和长期兴趣模型发送给成员引擎,从而使得成员引擎在获取搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化评分处理,从而获取各搜索结果信息相应的评分信息。当搜索服务器接收到成员引擎反馈的搜索结果信息以及相应的评分信息以后,还可以结合其它相关因素对搜索结果信息进行重新评分排序,从而可以获取个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can send the immediate interest model and long-term interest model extracted by the search application server to the member engine, so that after the member engine obtains the search result information, it can use the user's immediate interest model and long-term interest model to the member engine. The search result information is subjected to personalized scoring processing, so as to obtain the corresponding scoring information of each search result information. After the search server receives the search result information fed back by the member engine and the corresponding scoring information, it can also re-score and sort the search result information in combination with other relevant factors, so as to obtain personalized, satisfying user needs and high matching accuracy. Search result information.

图4为本发明移动搜索方法实施例四的流程图,如图4所示,本实施例的方法可以包括:FIG. 4 is a flow chart of Embodiment 4 of the mobile search method of the present invention. As shown in FIG. 4, the method of this embodiment may include:

步骤401、接收搜索服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。Step 401: Receive a search request sent by a search server, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server.

成员引擎可以接受搜索服务器发送的搜索请求。该成员引擎可以为采用实施例一和实施例二所述的方法选中的成员引擎。Membership engines can accept search requests sent by search servers. The member engine may be the member engine selected by the methods described in Embodiment 1 and Embodiment 2.

步骤402、对所述关键词信息进行搜索,获取搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理。Step 402: Search the keyword information, obtain search result information, and perform scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model.

成员引擎在接收该搜索请求后,可以对所需搜索的关键词信息进行搜索,获取搜索结果信息。然后,成员引擎可以根据即时兴趣模型和长期兴趣模型对搜索结果信息进行个性化评分处理,从而获取与各搜索结果信息对应的评分信息,该评分信息可以表示各搜索结果信息与即时兴趣模型和长期兴趣模型之间的匹配程度。After receiving the search request, the member engine can search for the keyword information to be searched to obtain search result information. Then, the member engine can perform personalized scoring processing on the search result information according to the instant interest model and the long-term interest model, so as to obtain the scoring information corresponding to each search result information. The degree of matching between the models of interest.

步骤403、将评分排序处理后的搜索结果信息反馈给所述搜索服务器。Step 403 , feeding back the search result information after scoring and sorting to the search server.

成员引擎可以将该搜索结果信息以及相应的评分信息发送给搜索服务器。The member engine can send the search result information and corresponding scoring information to the search server.

搜索服务器在接收到各成员引擎反馈的搜索结果信息以及相应的评分信息后,可以对搜索结果信息进行重新评分处理,即对该搜索结果信息进行进一步的筛选处理,从而可以获取更为个性化地,符合需求的搜索结果信息。在重新评分处理过程中,搜索服务器可以将成员引擎反馈的与各搜索结果信息对应的评分信息以及其他相关因素信息结合起来对搜索结果信息进行综合评分。该相关因素信息可以包括该成员引擎的价格、级别、搜索速度、好评率等等信息,本领域技术人员可以根据需要将各种可能影响排序的信息结合到重评分处理的过程中。After the search server receives the search result information and the corresponding scoring information fed back by each member engine, it can re-score the search result information, that is, further filter the search result information, so as to obtain more personalized results. , the search result information that meets the requirements. During the re-scoring process, the search server may combine the scoring information corresponding to each search result information fed back by the member engine and other relevant factor information to give a comprehensive score to the search result information. The relevant factor information may include information such as the member engine's price, level, search speed, favorable rate, etc. Those skilled in the art may combine various information that may affect ranking into the re-scoring process as required.

搜索服务器在完成重新评分处理后,即可将该重新评分处理后的搜索结果信息反馈给搜索应用服务器,以便搜索应用服务器可以将个性化的,满足用户需求的以及匹配精度高的搜索结果信息提供给搜索客户端。After the search server completes the re-scoring process, it can feed back the re-scored search result information to the search application server, so that the search application server can provide personalized search result information that meets user needs and has high matching accuracy. to the search client.

进一步地,步骤402中所述的根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理,可以包括:计算所述即时兴趣模型与所述搜索结果信息的第一相似度;计算所述长期兴趣模型与所述搜索结果信息的第二相似度;根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。Further, the scoring and sorting processing of the search result information according to the immediate interest model and the long-term interest model in step 402 may include: calculating the first similarity between the immediate interest model and the search result information ; Calculate the second similarity between the long-term interest model and the search result information; obtain scoring values according to the first similarity and second similarity, and sort the search result information according to the scoring values .

具体来说,搜索结果信息的评分值=r′1×搜索结果信息与LIM的第二相似度+r′2×搜索结果信息与IIM的第一相似度,r′1+r′2=1。Specifically, the score value of the search result information=r′ 1 ×the second similarity between the search result information and the LIM+r′ 2 ×the first similarity between the search result information and the IIM, r′ 1 +r′ 2 =1 .

计算搜索结果信息与长期兴趣模型向量LIM(r1,...,rn)的第二相似度:Calculate the second similarity between the search result information and the long-term interest model vector LIM(r1,...,rn):

(1)成员引擎根据倒排索引检索出候选的搜索结果信息。(1) The member engine retrieves candidate search result information according to the inverted index.

(2)成员引擎根据长期兴趣模型数据对候选的搜索结果信息进行个性化相关性评分。(2) The member engine performs personalized correlation scoring on the candidate search result information according to the long-term interest model data.

W=(r1,r2,r3,......,rn)为搜索服务器传过来的长期兴趣模型,D=(t1,t2,t3,.......,tn)为搜索结果信息所对应的兴趣模型向量。W=(r1, r2, r3,...,rn) is the long-term interest model sent by the search server, D=(t1, t2, t3,...,tn) is the search result The interest model vector corresponding to the information.

评分值score1=W×D=r1×t1+r2×t2+r3×t3+......,+rn×tn。Score value score1=W×D=r1×t1+r2×t2+r3×t3+...,+rn×tn.

计算搜索结果信息与即时兴趣模型IIM(u1,...,un)的第一相似度:Calculate the first similarity between the search result information and the instant interest model IIM(u1,...,un):

(1)成员引擎根据倒排索引检索出候选的搜索结果信息。(1) The member engine retrieves candidate search result information according to the inverted index.

(2)成员引擎根据即时兴趣模型数据对候选的搜索结果信息进行个性化相关性评分。(2) The member engine performs personalized correlation scoring on the candidate search result information according to the instant interest model data.

U=(u1,u2,u3,......,un)为搜索服务器传过来的即时兴趣模型,D=(t1,t2,t3,.......,tn)为搜索结果信息所对应的兴趣模型向量。U=(u1, u2, u3,..., un) is the instant interest model sent by the search server, D=(t1, t2, t3,..., tn) is the search result The interest model vector corresponding to the message.

评分值score2=W×D=u1×t1+u2×t2+u3×t3+......,+un×tnScore value score2=W×D=u1×t1+u2×t2+u3×t3+......, +un×tn

计算评分值=r′1×score1+r′2×score2。Calculation score = r' 1 ×score1+r' 2 ×score2.

本实施例中,成员引擎可以接收搜索服务器发送的即时兴趣模型和长期兴趣模型,从而在获取与关键词信息对应的搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化的评分处理,从而获取各搜索结果信息相应的个性化的评分信息,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the member engine can receive the immediate interest model and the long-term interest model sent by the search server, so that after obtaining the search result information corresponding to the keyword information, the search result can be searched according to the user's immediate interest model and long-term interest model. Personalized scoring processing is performed on the information, so as to obtain personalized scoring information corresponding to each search result information, and search result information that meets user needs and has high matching accuracy.

下面以一个具体实施例对本发明移动搜索方法上述实施例的技术方案进行详细说明。The technical solutions of the above embodiments of the mobile search method of the present invention will be described in detail below with a specific embodiment.

图5为本发明移动搜索方法实施例五的信令流程图,如图5所示,本实施例的方法可以包括:FIG. 5 is a signaling flow chart of Embodiment 5 of the mobile search method of the present invention. As shown in FIG. 5, the method of this embodiment may include:

步骤501、搜索客户端将搜索请求发给搜索应用服务器。Step 501, the search client sends a search request to the search application server.

该搜索请求中可以携带所需搜索的关键词信息。The search request may carry keyword information to be searched.

步骤502、搜索应用服务器从用户数据库中提取用户的长期兴趣模型和即时兴趣模型。Step 502, the search application server extracts the user's long-term interest model and immediate interest model from the user database.

例如,搜索应用服务器可以从用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型。For example, the search application server can extract the user's long-term interest model from the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract the user's long-term interest model from the current query q( t) Extract the user's instant interest model from the relevant data of the query sequence q(1), ..., q(t-1), q(t) in the same search context session.

步骤503、搜索应用服务器向搜索服务器发送搜索请求。Step 503, the search application server sends a search request to the search server.

该搜索请求中可以携带关键词信息以及用户的长期兴趣模型和即时兴趣模型。The search request may carry keyword information and the user's long-term interest model and immediate interest model.

步骤504、搜索服务器向成员引擎分发搜索请求。Step 504, the search server distributes search requests to member engines.

该搜索请求中可以携带关键词信息以及用户的长期兴趣模型和即时兴趣模型。成员引擎可以接受搜索服务器发送的搜索请求。该成员引擎可以为采用实施例一和实施例二所述的方法选中的成员引擎。The search request may carry keyword information and the user's long-term interest model and immediate interest model. Membership engines can accept search requests sent by search servers. The member engine may be the member engine selected by the methods described in Embodiment 1 and Embodiment 2.

步骤505、成员引擎搜索关键词信息,获取搜索结果信息,根据用户的即时兴趣模型和长期兴趣模型对搜索结果信息进行个性化相关性评分与排序处理。Step 505 , the member engine searches keyword information, obtains search result information, and performs personalized correlation scoring and sorting processing on the search result information according to the user's immediate interest model and long-term interest model.

获取评分值的过程可以采用上述方法,不再赘述。The above-mentioned method may be used for the process of obtaining the scoring value, and details are not repeated here.

步骤506、成员引擎将排序处理后的搜索结果信息以及相应的评分值返回给搜索服务器。Step 506, the member engine returns the sorted search result information and the corresponding scoring value to the search server.

步骤507、搜索服务器根据搜索结果信息的评分值和其它相关因素对搜索结果信息进行重新评分。Step 507, the search server re-scores the search result information according to the score value of the search result information and other relevant factors.

例如,搜索结果信息的综合评分值=r1×成员引擎返回评分值+r2×成员引擎级别相关的评分值+r3×价格因素相关的评分值,其中r1+r2+r3=1。For example, the comprehensive score value of search result information=r1*return score value of member engine+r2*score value related to member engine level+r3*score value related to price factor, where r1+r2+r3=1.

步骤508、搜索服务器根据综合评分值,对搜索结果信息进行重新排序处理。Step 508, the search server reorders the search result information according to the comprehensive scoring value.

步骤509、搜索服务器将最终的搜索结果信息发送给搜索应用服务器。Step 509, the search server sends the final search result information to the search application server.

步骤510、搜索应用服务器将最终的搜索结果信息发送给搜索客户端。Step 510, the search application server sends the final search result information to the search client.

本实施例中,搜索服务器可以将搜索应用服务器提取的即时兴趣模型和长期兴趣模型发送给成员引擎,从而使得成员引擎在获取搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化的评分处理,从而获取各搜索结果信息相应的评分信息。当搜索服务器接收到成员引擎反馈的搜索结果信息以及相应的评分信息以后,还可以结合其它相关因素对搜索结果信息进行重新评分排序,从而可以获取个性化的、满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can send the immediate interest model and long-term interest model extracted by the search application server to the member engine, so that after the member engine obtains the search result information, it can use the user's immediate interest model and long-term interest model to the member engine. The search result information is subjected to personalized scoring processing, so as to obtain the corresponding scoring information of each search result information. After the search server receives the search result information fed back by the member engine and the corresponding scoring information, it can also re-score and sort the search result information in combination with other relevant factors, so as to obtain personalized, satisfying user needs and high matching accuracy. Search result information.

图6为本发明移动搜索方法实施例六的流程图,如图6所示,本实施例的方法可以包括:FIG. 6 is a flow chart of Embodiment 6 of the mobile search method of the present invention. As shown in FIG. 6, the method of this embodiment may include:

步骤601、接收搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。Step 601: Receive a search request sent by a search application server, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server.

举例来说,搜索服务器可以接收搜索应用服务器发送的搜索请求。该搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型。搜索应用服务器可以根据搜索客户端发送的搜索请求消息,从用户数据库中获取用户的即时兴趣模型和长期兴趣模型。例如,从用户数据库,如用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型。在提取即时兴趣模型和长期兴趣模型后,搜索应用服务器即可向搜索服务器发送搜索请求,从而使得搜索服务器可以根据该搜索请求中携带的用户的即时兴趣模型和长期兴趣模型对所需搜索的关键字信息进行搜索。For example, the search server may receive the search request sent by the search application server. The search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server. The search application server can obtain the user's immediate interest model and long-term interest model from the user database according to the search request message sent by the search client. For example, extract the user's long-term interest model from the user database, such as the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract from the current query q( t) Extract the user's instant interest model from the relevant data of the query sequence q(1), ..., q(t-1), q(t) in the same search context session. After extracting the immediate interest model and the long-term interest model, the search application server can send a search request to the search server, so that the search server can determine the key information of the desired search according to the user's immediate interest model and long-term interest model carried in the search request. word information to search.

用户的兴趣模型可以用n个维度来表示如:新闻、体育、娱乐、财经、科技、房产、游戏、女性、论坛、天气、商品、家电、音乐、读书、博客、手机、军事、教育、旅游、彩信、彩铃、餐饮、民航、工业、农业、电脑、地理等。用户对每个维度的兴趣的评分值所组成的一个向量W(r1,r2,r3,......,rn)则为用户的兴趣模型。The user's interest model can be represented by n dimensions such as: news, sports, entertainment, finance, technology, real estate, games, women, forums, weather, commodities, home appliances, music, reading, blogs, mobile phones, military, education, tourism , MMS, CRBT, catering, civil aviation, industry, agriculture, computer, geography, etc. A vector W(r1, r2, r3, . . . , rn) composed of rating values of the user's interest in each dimension is the user's interest model.

如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由用户的所有搜索历史数据和用户的静态档案profile计算得到,则兴趣模型W(r1,r2,r3,...rn)为用户的长期兴趣模型。如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到的,那么W(r1,r2,r3,...rn)为用户的即时兴趣模型。If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is calculated from all the user's search history data and the user's static profile profile, then the interest model W(r1, r2 , r3,...rn) is the user's long-term interest model. If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is determined by the query sequence q(1) in the same search context session as the current query q(t), ... , q(t-1), q(t) related data is calculated, then W(r1, r2, r3,...rn) is the user's instant interest model.

步骤602、接收成员引擎根据所述关键词信息搜索获取的搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理。Step 602: Receive the search result information acquired by the member engine according to the keyword information, and perform scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model.

成员引擎可以根据搜索服务器发送的搜索请求中携带的关键词信息进行搜索,获取搜索结果信息,并将该搜索结果信息反馈给搜索服务器。搜索服务器可以根据即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化的评分排序处理。The member engine can search according to the keyword information carried in the search request sent by the search server, obtain search result information, and feed back the search result information to the search server. The search server can perform personalized scoring and ranking processing on the search result information according to the immediate interest model and the long-term interest model.

步骤603、将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述搜索应用服务器。Step 603 , sending the search result information after scoring and ranking processing and the corresponding scoring information to the search application server.

搜索服务器可以将评分排序处理后的搜索结果信息以及相应的评分信息发送给搜索应用服务器,以便搜索应用服务器将搜索结果信息以及相应的评分信息反馈给搜索客户端,从而为用户提供个性化的,精度较高的搜索结果。The search server can send the search result information and the corresponding scoring information after scoring sorting to the search application server, so that the search application server can feed back the search result information and the corresponding scoring information to the search client, thereby providing users with personalized, Higher precision search results.

进一步地,步骤602所述的根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理,可以包括:计算所述即时兴趣模型与所述搜索结果信息的第一相似度;计算所述长期兴趣模型与所述搜索结果信息的第二相似度;根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。Further, the scoring and sorting processing of the search result information according to the immediate interest model and the long-term interest model in step 602 may include: calculating a first similarity between the immediate interest model and the search result information; Calculating a second similarity between the long-term interest model and the search result information; obtaining a score value according to the first similarity and the second similarity, and sorting the search result information according to the score value.

具体来说,搜索结果信息的评分值=r′1×搜索结果信息与LIM的第二相似度+r′2×搜索结果信息与IIM的第一相似度,r′1+r′2=1。Specifically, the score value of the search result information=r′ 1 ×the second similarity between the search result information and the LIM+r′ 2 ×the first similarity between the search result information and the IIM, r′ 1 +r′ 2 =1 .

计算搜索结果信息与长期兴趣模型向量LIM(r1,...,rn)的第二相似度:Calculate the second similarity between the search result information and the long-term interest model vector LIM(r1,...,rn):

(1)成员引擎根据倒排索引检索出候选的搜索结果信息。(1) The member engine retrieves candidate search result information according to the inverted index.

(2)成员引擎根据长期兴趣模型数据对候选的搜索结果信息进行个性化相关性评分。(2) The member engine performs personalized correlation scoring on the candidate search result information according to the long-term interest model data.

W=(r1,r2,r3,......,rn)为搜索服务器传过来的长期兴趣模型,D=(t1,t2,t3,.......,tn)为搜索结果信息所对应的兴趣模型向量。W=(r1, r2, r3,...,rn) is the long-term interest model sent by the search server, D=(t1, t2, t3,...,tn) is the search result The interest model vector corresponding to the message.

评分值score1=W×D=r1×t1+r2×t2+r3×t3+......,+rn×tn。Score value score1=W×D=r1×t1+r2×t2+r3×t3+...,+rn×tn.

计算搜索结果信息与即时兴趣模型IIM(u1,...,un)的第一相似度:Calculate the first similarity between the search result information and the instant interest model IIM(u1,...,un):

(1)成员引擎根据倒排索引检索出候选的搜索结果信息。(1) The member engine retrieves candidate search result information according to the inverted index.

(2)成员引擎根据即时兴趣模型数据对候选的搜索结果信息进行个性化相关性评分。(2) The member engine performs personalized correlation scoring on the candidate search result information according to the instant interest model data.

U=(u1,u2,u3,......,un)为搜索服务器传过来的即时兴趣模型,D=(t1,t2,t3,.......,tn)为搜索结果信息所对应的兴趣模型向量。U=(u1, u2, u3,..., un) is the instant interest model sent by the search server, D=(t1, t2, t3,..., tn) is the search result The interest model vector corresponding to the information.

评分值score2=W×D=u1×t1+u2×t2+u3×t3+......,+un×tnScore value score2=W×D=u1×t1+u2×t2+u3×t3+......, +un×tn

计算评分值=r′1×score1+r′2×score2。Calculation score = r' 1 ×score1+r' 2 ×score2.

搜索服务器即可根据评分值对搜索结果信息进行排序处理,获取排序处理后的搜索结果信息。The search server can sort the search result information according to the scoring value, and obtain the sorted search result information.

本实施例中,搜索服务器可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎反馈的搜索结果信息进行评分排序,从而用户提供个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can rank and rank the search result information fed back by the member engine according to the user's immediate interest model and long-term interest model, so that the user can provide personalized search result information that meets the user's needs and has high matching accuracy.

下面以一个具体实施例对本发明移动搜索方法上述实施例的技术方案进行详细说明。The technical solutions of the above embodiments of the mobile search method of the present invention will be described in detail below with a specific embodiment.

图7为本发明移动搜索方法实施例七的信令流程图,如图7所示,本实施例的方法可以包括:FIG. 7 is a signaling flow chart of Embodiment 7 of the mobile search method of the present invention. As shown in FIG. 7, the method of this embodiment may include:

步骤701、搜索客户端将搜索请求发给搜索应用服务器。Step 701, the search client sends a search request to the search application server.

该搜索请求中可以携带所需搜索的关键词信息。The search request may carry keyword information to be searched.

步骤702、搜索应用服务器从用户数据库中提取用户的长期兴趣模型和即时兴趣模型。Step 702, the search application server extracts the user's long-term interest model and immediate interest model from the user database.

例如,搜索应用服务器可以从用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型。For example, the search application server can extract the user's long-term interest model from the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract the user's long-term interest model from the current query q( t) Extract the user's instant interest model from the relevant data of the query sequence q(1), ..., q(t-1), q(t) in the same search context session.

步骤703、搜索应用服务器向搜索服务器发送搜索请求。Step 703, the search application server sends a search request to the search server.

该搜索请求中可以携带关键词信息以及用户的长期兴趣模型和即时兴趣模型。The search request may carry keyword information and the user's long-term interest model and immediate interest model.

步骤704、搜索服务器向成员引擎分发搜索请求。Step 704, the search server distributes search requests to member engines.

该搜索请求中可以携带关键词信息。成员引擎可以接受搜索服务器发送的搜索请求。该成员引擎可以为采用实施例一和实施例二所述的方法选中的成员引擎。The search request may carry keyword information. Membership engines can accept search requests sent by search servers. The member engine may be the member engine selected by the methods described in Embodiment 1 and Embodiment 2.

步骤705、各个成员引擎完成搜索,获取搜索结果信息。Step 705, each member engine completes the search and obtains search result information.

步骤706、各个成员引擎将搜索结果信息返回给搜索服务器。Step 706, each member engine returns the search result information to the search server.

步骤707、搜索服务器根据即时兴趣模型和长期兴趣模型对搜索结果信息进行评分排序处理。Step 707, the search server performs scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model.

评分排序处理的过程如上所述,不再赘述。The process of scoring and sorting is as described above and will not be repeated here.

步骤708、搜索服务器将评分排序处理后的搜索结果信息返回给搜索应用服务器。Step 708, the search server returns the search result information after scoring and sorting to the search application server.

步骤709、搜索应用服务器将最终的搜索结果信息返回给搜索客户端。Step 709, the search application server returns the final search result information to the search client.

本实施例中,搜索服务器可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎反馈的搜索结果信息进行评分排序,从而用户提供个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can rank and rank the search result information fed back by the member engine according to the user's immediate interest model and long-term interest model, so that the user can provide personalized search result information that meets the user's needs and has high matching accuracy.

图8为本发明移动搜索方法实施例八的流程图,如图8所示,本实施例的方法可以包括:FIG. 8 is a flow chart of Embodiment 8 of the mobile search method of the present invention. As shown in FIG. 8, the method of this embodiment may include:

步骤801、接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息。Step 801. Receive a search request message sent by a search client, where the search request message carries keyword information.

具体来说,搜索应用服务器可以接收搜索客户端发送的搜索请求消息,该搜索请求消息中可以携带所需搜索的关键词信息。Specifically, the search application server may receive a search request message sent by the search client, and the search request message may carry keyword information to be searched.

步骤802、从用户数据库中提取即时兴趣模型和长期兴趣模型。Step 802, extract the immediate interest model and long-term interest model from the user database.

搜索应用服务器可以从用户数据库中提取即时兴趣模型和长期兴趣模型。例如,搜索应用服务器可以从用户的静态profile、搜索历史等信息中提取用户的长期兴趣模型,或者直接提取预先存储在用户数据库中的长期兴趣模型,并且搜索应用服务器还可以从与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据中提取用户的即时兴趣模型。The search application server can extract the immediate interest model and the long-term interest model from the user database. For example, the search application server can extract the user's long-term interest model from the user's static profile, search history and other information, or directly extract the long-term interest model pre-stored in the user database, and the search application server can also extract the user's long-term interest model from the current query q( t) Extract the user's instant interest model from the relevant data of the query sequence q(1), ..., q(t-1), q(t) in the same search context session.

举例来说,用户的兴趣模型可以用n个维度来表示如:新闻、体育、娱乐、财经、科技、房产、游戏、女性、论坛、天气、商品、家电、音乐、读书、博客、手机、军事、教育、旅游、彩信、彩铃、餐饮、民航、工业、农业、电脑、地理等。用户对每个维度的兴趣的评分值所组成的一个向量W(r1,r2,r3,......,rn)则为用户的兴趣模型。For example, the user's interest model can be represented by n dimensions such as: news, sports, entertainment, finance, technology, real estate, games, women, forums, weather, commodities, home appliances, music, reading, blogs, mobile phones, military , education, tourism, MMS, CRBT, catering, civil aviation, industry, agriculture, computer, geography, etc. A vector W(r1, r2, r3, . . . , rn) composed of rating values of the user's interest in each dimension is the user's interest model.

如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由用户的所有搜索历史数据和用户的静态档案profile计算得到,则兴趣模型W(r1,r2,r3,...rn)为用户的长期兴趣模型。如果兴趣模型W(r1,r2,r3,...,rn)中的各个维度的评分值ri是由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到的,那么W(r1,r2,r3,...rn)为用户的即时兴趣模型。If the rating value ri of each dimension in the interest model W(r1, r2, r3,...,rn) is calculated from all the user’s search history data and the user’s static profile profile, then the interest model W(r1, r2 , r3,...rn) is the user's long-term interest model. If the rating value ri of each dimension in the interest model W(r1, r2, r3, ..., rn) is derived from the query sequence q(1) in the same search context session as the current query q(t), ... , q(t-1), q(t) related data is calculated, then W(r1, r2, r3,...rn) is the user's instant interest model.

对于用户的静态profile所对应的长期兴趣模型W1来说,W1=(p1,p2,p3,......,pn),其中pi为静态profile中类型属于第i个兴趣维度的所有词的词频之和;或者对该静态profile对应的文档进行分类(分类的算法可以用朴素贝叶斯、K最近邻分类算法、支持向量机、向量空间模型等算法),tj等于该静态profile对应的文档属于第j个兴趣维度所对应的类型的评分值。For the long-term interest model W1 corresponding to the user's static profile, W1=(p1, p2, p3, ..., pn), where pi is all the words in the static profile that belong to the i-th interest dimension or classify the documents corresponding to the static profile (the classification algorithm can use Naive Bayesian, K nearest neighbor classification algorithm, support vector machine, vector space model and other algorithms), tj is equal to the static profile corresponding to The document belongs to the rating value of the type corresponding to the jth interest dimension.

对于用户的搜索点击历史所对应的长期兴趣模型W2来说,W2=d1+d2+d3+......dm,其中di为用户某个点击文档所对应的兴趣模型向量,di=(t1,t2,t3,.......,tn),当用户最新点击了这个文档,tj等于文档中类型属于第j个兴趣维度的所有词的词频之和;或者对该文档进行分类(分类的算法可以用朴素贝叶斯、K最近邻分类算法、支持向量机、向量空间模型等算法),tj等于该文档属于第j个兴趣维度所对应的类型的评分值。如果用户对某个点击过的文档进行评价,如果评价好,di向量乘以一个正的常数c表示文档的重要性增加di=c×di=(c×ti,c×t2,c×t3,......,c×tn),如果评价不好,di向量乘以一个正的常数c的倒数表示文档的重要性减小di=1/c×di=(1/c×ti,1/c×t2,1/c×t3,......,1/c×tn);过了一段时间,tj的值又自动减少一定的百分比,表示随着时间的推移其重要性减弱,直到过了较长的时间tj的值减为零为止,这时将di从历史记录中删除。For the long-term interest model W2 corresponding to the user's search click history, W2=d1+d2+d3+...dm, where di is the interest model vector corresponding to a document clicked by the user, di=(t1 , t2, t3,......, tn), when the user clicks on this document recently, tj is equal to the sum of the word frequencies of all words in the document whose type belongs to the jth dimension of interest; or classify the document ( The classification algorithm can use naive Bayesian, K nearest neighbor classification algorithm, support vector machine, vector space model and other algorithms), and tj is equal to the score value of the type corresponding to the jth dimension of interest that the document belongs to. If the user evaluates a document that has been clicked, if the evaluation is good, the di vector is multiplied by a positive constant c to indicate that the importance of the document increases di=c×di=(c×ti, c×t2, c×t3, ......, c×tn), if the evaluation is not good, the reciprocal of the di vector multiplied by a positive constant c indicates that the importance of the document is reduced di=1/c×di=(1/c×ti, 1/c×t2, 1/c×t3, ..., 1/c×tn); after a period of time, the value of tj will automatically decrease by a certain percentage, indicating its importance over time Weaken until the value of tj decreases to zero after a long period of time, then delete di from the history.

对于综合的长期兴趣模型来说,可以将W1和W2分别归一化后相加,即:兴趣模型向量W=W1+W2,或者加权相加,如兴趣模型向量W=W1×30%+W2×70%,然后再对W进行归一化处理。For a comprehensive long-term interest model, W1 and W2 can be normalized and added separately, that is: interest model vector W=W1+W2, or weighted addition, such as interest model vector W=W1×30%+W2 ×70%, and then normalize W.

步骤803、向搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型,以使所述搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。Step 803: Send a search request to the search server, the search request carries keyword information and the immediate interest model and long-term interest model, so that the search server can search for the keyword according to the immediate interest model and long-term interest model word information to search.

搜索应用服务器可以将关键词信息发送给搜索服务器,也将提取获得的即时兴趣模型和长期兴趣模型也发送给搜索服务器,从而使得搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。搜索服务器具体进行搜索的过程可以采用上述实施例三~七中的方法实现。The search application server can send the keyword information to the search server, and also send the extracted immediate interest model and long-term interest model to the search server, so that the search server can search for the keyword according to the immediate interest model and long-term interest model. information to search. The specific search process of the search server can be realized by using the methods in the third to seventh embodiments above.

进一步地,步骤802中所述的从用户数据库中提取即时兴趣模型,可以包括:应用条件随机场模型计算在给定与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),....q(t-1),q(t)的条件下,当前查询q(t)的输出类型的条件概率,将该条件概率值作为即时兴趣模型与该输出类型对应的兴趣维度的评分值。Further, extracting the instant interest model from the user database in step 802 may include: applying the conditional random field model to calculate the query sequence q(1) in a given session in the same search context as the current query q(t), ....q(t-1), under the condition of q(t), the conditional probability of the output type of the current query q(t), the conditional probability value is used as the instant interest model and the corresponding dimension of interest of the output type Rating value.

举例来说,本实施例可以定义G=(V,E)为一个无向图,Y={Yv|v∈V}.即V中的每个节点对应一个随机变量所表示的标记序列的成分Yv,如果每个随机量Yv对于G遵守马尔可夫属性,那么(X,Y)就是一个条件随机场,而且在给定X和所有其他随机变量Y{u|u≠v,{u,v}∈V}的条件下,随机变量Yv的概率P(Yv|X,Yu,u≠v,{u.v}∈V)即等于P(Yv|X,Yu,(u,v)∈E)。For example, this embodiment can define G=(V, E) as an undirected graph, Y={Y v |v∈V}. That is, each node in V corresponds to a sequence of labels represented by a random variable component Y v , if each random quantity Y v obeys the Markov property for G, then (X, Y) is a conditional random field, and given X and all other random variables Y {u|u≠v, { Under the condition of u, v}∈V} , the probability P(Y v |X, Y u , u≠v, {uv}∈V) of random variable Y v is equal to P(Y v |X, Y u , ( u, v) ∈ E).

根据马尔可夫属性和最大熵的原理,可以推导出条件随机场的经典条件概率公式:According to the Markov property and the principle of maximum entropy, the classical conditional probability formula of the conditional random field can be derived:

给定观察序列x的前提下,观察序列的状态标记序列y的条件概率等于:Given the observation sequence x, the conditional probability of the state label sequence y of the observation sequence is equal to:

PP &theta;&theta; (( ythe y || xx )) == 11 ZZ (( xx )) expexp (( &Sigma;&Sigma; ee &Element;&Element; EE. ,, kk &lambda;&lambda; kk ff kk (( ee ,, ythe y || ee ,, xx )) ++ &Sigma;&Sigma; vv &Element;&Element; VV ,, kk uu kk gg kk (( vv ,, ythe y || vv ,, xx )) ))

其中,x为观察序列,y为标记状态序列,y|S为与子图S的顶点相关联的序列y的成分的集合,f,g为特征函数,λ,μ为特征函数的权重值,Z(x)为归一化因子。Among them, x is the observation sequence, y is the marked state sequence, y| S is the set of components of the sequence y associated with the vertices of the subgraph S, f, g are the characteristic functions, λ, μ are the weight values of the characteristic functions, Z(x) is a normalization factor.

给定处于同一session的查询序列q=q1,...,q(T-1),q(T),输出查询序列对应的类型序列C=c1,...cT-1,cT的条件概率:Given a query sequence q=q1,...,q(T-1), q(T) in the same session, output the type sequence C=c 1 ,...c T-1 , c T corresponding to the query sequence The conditional probability of :

令ci的取值空间为|C|,令c0=start,cT+1=end,将状态start、end加入|C|Let the value space of c i be |C|, let c 0 = start, c T+1 = end, add the states start and end to |C|

pp (( cc || qq )) == 11 ZZ (( qq )) &Pi;&Pi; tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) ,,

其中: Z ( q ) = &Sigma; c &Pi; t = 1 T + 1 M t ( c t - 1 , c t | q ) , 为归一化因子。in: Z ( q ) = &Sigma; c &Pi; t = 1 T + 1 m t ( c t - 1 , c t | q ) , is the normalization factor.

Mm tt (( cc tt -- 11 ,, cc tt || qq )) == expexp (( &Sigma;&Sigma; kk &lambda;&lambda; kk ff kk (( cc tt -- 11 ,, cc tt ,, qq )) ++ &Sigma;&Sigma; kk uu kk gg kk (( cc tt ,, qq )) ))

定义一个|C|×|C|的矩阵:Define a |C|×|C| matrix:

Mt(q)=[Mt(ct-1,ct|q)]M t (q)=[M t (c t-1 , c t |q)]

那么Z(q)等于M1(q)*...MT(q)*MT+1(q)矩阵的(start,end)项。Then Z(q) is equal to the (start, end) items of the M 1 (q)*...M T (q)*M T+1 (q) matrix.

ZZ (( qq )) == &Sigma;&Sigma; cc &Pi;&Pi; tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) == &Sigma;&Sigma; startcstartc 11 .. .. .. cc TT -- 11 cc TT endend &Pi;&Pi; tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq ))

== (( &Pi;&Pi; tt == 11 TT ++ 11 Mm tt (( qq )) )) startstart ,, endend

令θ=(λ1,.λ2,......;u1,u2,......)Let θ=(λ 1 ,.λ 2 , . . . ; u 1 , u 2 , . . . )

参数θ的确定:Determination of parameter θ:

给定训练数据

Figure B2009102208107D0000387
和经验分布
Figure B2009102208107D0000388
given training data
Figure B2009102208107D0000387
and experience distribution
Figure B2009102208107D0000388

训练数据的对数似然函数为:The log-likelihood function of the training data is:

LL (( &theta;&theta; )) == &Sigma;&Sigma; ii == 11 NN loglog PP &theta;&theta; (( cc (( ii )) || qq (( ii )) ))

&Proportional;&Proportional; &Sigma;&Sigma; qq ,, cc pp ~~ (( qq ,, cc )) loglog (( pp &theta;&theta; (( cc || qq )) ))

求θ使得L(θ)取得最大值。Find θ such that L(θ) takes the maximum value.

本实施例可以用GIS算法求θ:In this embodiment, GIS algorithm can be used to find θ:

(a)求Efk、Egk(a) Find Ef k , Eg k :

EfEf kk == &Sigma;&Sigma; qq ,, cc pp &theta;&theta; (( qq ,, cc )) ff kk (( qq ,, cc ))

== &Sigma;&Sigma; qq ,, cc pp &theta;&theta; (( qq )) pp &theta;&theta; (( cc || qq )) ff kk (( qq ,, cc ))

&ap;&ap; &Sigma;&Sigma; qq ,, cc pp ~~ (( qq )) pp &theta;&theta; (( cc || qq )) ff kk (( qq ,, cc ))

== &Sigma;&Sigma; qq pp ~~ (( qq )) &Sigma;&Sigma; ii == 11 TT &Sigma;&Sigma; cc ii -- 11 cc ii (( pp &theta;&theta; (( cc ii -- 11 ,, cc ii || qq )) ff kk (( cc ii -- 11 ,, cc ii ,, qq )) ))

pp &theta;&theta; (( cc ii -- 11 ,, cc ii || qq )) == &Sigma;&Sigma; startcstartc 11 .. .. .. cc ii -- 22 cc ii ++ 11 .. .. .. cc TendTend pp &theta;&theta; (( startcstartc 11 ,, .. .. .. ,, cc TT endend || qq ))

== 11 ZZ (( qq )) &Sigma;&Sigma; startcstartc 11 .. .. .. cc ii -- 22 cc ii ++ 11 .. .. .. cc TendTend &Pi;&Pi; tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq ))

== 11 ZZ (( qq )) (( &Sigma;&Sigma; startcstartc 11 .. .. .. cc ii -- 22 &Pi;&Pi; tt == 11 ii -- 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) )) Mm ii (( cc ii -- 11 ,, cc ii || qq )) (( &Sigma;&Sigma; CC ii ++ 11 .. .. .. CC TT endend &Pi;&Pi; tt == ii ++ 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) ))

== 11 ZZ (( qq )) (( (( &Pi;&Pi; tt == 11 ii -- 11 Mm tt (( qq )) )) )) startstart ,, cc ii -- 11 Mm ii (( cc ii -- 11 ,, cc ii || qq )) (( &Pi;&Pi; tt == ii ++ 11 TT ++ 11 Mm tt (( qq )) )) cc ii ,, endend ))

== 11 ZZ (( qq )) &alpha;&alpha; ii -- 11 (( cc ii -- 11 || qq )) Mm ii (( cc ii -- 11 ,, cc ii || qq )) &beta;&beta; ii (( cc ii || qq )) ))

其中:in:

αi(q)为1×|C|向量,α i (q) is a 1×|C| vector,

αi(q)=αi-1Mi(q)α i (q) = α i-1 M i (q)

如果ct=start,α0(ct|q)=1If c t =start, α 0 (c t |q)=1

否则α0(ct|q)=0Otherwise α 0 (c t |q)=0

β(i)为1x|C|向量,β(i) is a 1x|C| vector,

β(i)T=Mi+1(q)β(i+1)T β(i) T =M i+1 (q)β(i+1) T

如果ct=endβT+1(ct|q)=1If c t =endβ T+1 (c t |q)=1

否则,βT+1(ct|q)=0Otherwise, β T+1 (c t |q)=0

Egeg kk == &Sigma;&Sigma; qq ,, cc pp &theta;&theta; (( qq ,, cc )) gg kk (( qq ,, cc ))

== &Sigma;&Sigma; qq ,, cc pp &theta;&theta; (( qq )) pp &theta;&theta; (( cc || qq )) gg kk (( qq ,, cc ))

&ap;&ap; &Sigma;&Sigma; qq ,, cc pp ~~ (( qq )) pp &theta;&theta; (( cc || qq )) gg kk (( qq ,, cc ))

== &Sigma;&Sigma; qq pp ~~ (( qq )) &Sigma;&Sigma; ii == 11 TT &Sigma;&Sigma; cc ii pp &theta;&theta; (( cc ii || qq )) gg kk (( cc ii ,, qq ))

pp &theta;&theta; (( cc ii || qq )) == &Sigma;&Sigma; cc 11 .. .. .. cc ii -- 11 cc ii ++ 11 .. .. .. cc TT pp &theta;&theta; (( cc 11 ,, .. .. .. ,, cc TT || qq ))

== 11 ZZ (( qq )) &Sigma;&Sigma; cc 11 .. .. .. cc ii -- 11 cc ii ++ 11 .. .. .. cc TT &Pi;&Pi; tt == 11 TT Mm tt (( cc tt -- 11 ,, cc tt ,, qq ))

== 11 ZZ (( qq )) (( &Sigma;&Sigma; cc 11 .. .. .. cc ii -- 11 &Pi;&Pi; tt == 11 ii Mm tt (( cc tt -- 11 ,, cc tt ,, qq )) )) (( &Sigma;&Sigma; cc ii ++ 11 .. .. .. cc TT &Pi;&Pi; tt == ii ++ 11 TT Mm tt (( cc tt -- 11 ,, cc tt ,, qq )) ))

== 11 ZZ (( qq )) (( (( &Pi;&Pi; tt == 11 ii Mm tt (( qq )) )) )) startstart ,, cc ii (( &Pi;&Pi; tt == ii ++ 11 TT ++ 11 Mm tt (( qq )) )) cc ii ,, endend ))

== 11 ZZ (( qq )) &alpha;&alpha; (( ii )) &beta;&beta; (( ii ))

(b)求 (b) seeking

EE. ~~ ff kk == &Sigma;&Sigma; qq ,, cc pp ~~ (( qq ,, cc )) ff kk (( qq ,, cc ))

== &Sigma;&Sigma; qq ,, cc pp ~~ (( qq )) pp ~~ (( cc || qq )) ff kk (( qq ,, cc ))

== &Sigma;&Sigma; qq pp ~~ (( qq )) &Sigma;&Sigma; ii == 11 TT &Sigma;&Sigma; cc ii -- 11 cc ii pp ~~ (( cc ii -- 11 ,, cc ii || qq )) ff kk (( cc ii -- 11 ,, cc ii ,, qq ))

EE. ~~ gg kk == &Sigma;&Sigma; qq ,, cc pp ~~ (( qq ,, cc )) gg kk (( qq ,, cc ))

== &Sigma;&Sigma; qq ,, cc pp ~~ (( qq )) pp ~~ (( cc || qq )) gg kk (( qq ,, cc ))

== &Sigma;&Sigma; qq pp ~~ (( qq )) &Sigma;&Sigma; ii == 11 TT &Sigma;&Sigma; cc ii pp ~~ (( cc ii || qq )) gg kk (( cc ii ,, qq ))

(c)求迭代求λk、uk,直到λk、uk收敛:(c) Calculate λ k and u k iteratively until λ k and u k converge:

Figure B2009102208107D0000414
其中S1为大于1的常数,使得对任何
Figure B2009102208107D0000415
Figure B2009102208107D0000414
where S 1 is a constant greater than 1, such that for any
Figure B2009102208107D0000415

λk+1=λk+δλk λ k+1 =λ k +δλ k

其中S2为大于1的常数,使得对任何q、c, &Sigma; k = 0 n g k ( q , c ) = S 2 where S 2 is a constant greater than 1, so that for any q, c, &Sigma; k = 0 no g k ( q , c ) = S 2

uk+1=uk+δuk u k+1 =u k +δu k

重复(a)、(b)、(c)步骤直到λk、uk收敛。Repeat steps (a), (b) and (c) until λ k and u k converge.

给定处于同一session查询序列q=q1,...,q(T-1),q(T),当前查询q(T)属于类型cT的条件概率:Given the query sequence q=q1,...,q(T-1), q(T) in the same session, the conditional probability that the current query q(T) belongs to type c T :

pp (( cc TT || qq )) == &Sigma;&Sigma; cc 11 .. .. .. cc TT -- 11 pp (( cc || qq ))

== 11 ZZ (( qq )) (( &Sigma;&Sigma; startcstartc 11 .. .. .. cc TT -- 11 &Pi;&Pi; tt == 11 TT ++ 11 Mm tt (( cc tt -- 11 ,, cc tt || qq )) ))

== 11 ZZ (( qq )) (( &Sigma;&Sigma; startcstartc 11 .. .. .. cc TT -- 11 &Pi;&Pi; tt == 11 TT Mm tt (( cc tt -- 11 ,, cc tt || qq )) )) Mm TT ++ 11 (( cc TT ,, cc endend || qq )) ))

== 11 ZZ (( qq )) (( (( &Pi;&Pi; tt == 11 TT Mm tt (( qq )) )) )) startstart ,, cc TT Mm TT ++ 11 (( cc TT ,, cc endend || qq ))

== 11 ZZ (( qq )) &alpha;&alpha; TT (( cc TT || qq )) Mm TT ++ 11 (( cc TT ,, cc endend || qq ))

把p(cT|q)作为即时兴趣模型类型为cT的对应维度的评分值。Take p(c T |q) as the score value of the corresponding dimension of the instant interest model type c T.

本地特征函数gk的选取:Selection of local feature function g k :

(1)给每个领域类型cT的所有主题词和相关词赋予一定的权重,由这些主题词和相关词的权重组成一个领域cT的向量,cT(t1,...,tn-1,tn)(1) Give certain weights to all subject words and related words of each field type c T , and form a vector of field c T by the weights of these subject words and related words, c T (t 1 ,...,t n-1 , t n )

cT中的词的权重的分配方法有两种,There are two ways to assign the weight of words in c T ,

一种是人工分配权重的方法:One is the method of manually assigning weights:

cT的词的权重可以这样赋予:对于主题词赋予最大的权重,对于强相关词赋予中间大小的权重,对于弱相关词赋予最小权重。The weight of the words of c T can be given as follows: give the largest weight to the subject words, give the middle weight to the strong related words, and give the smallest weight to the weak related words.

比如:主题词(如餐饮领域cT的”川菜”)赋予权重1,强相关词(如餐饮领域cT的”辣”)赋予权重0.8,弱相关词(如餐饮领域cT的”香”)赋予权重0.5For example: subject words (such as "Sichuan Cuisine" in the catering field c T ) are given a weight of 1, strong related words (such as "spicy" in the catering field c T ) are given a weight of 0.8, weakly related words (such as "fragrant" in the catering field c T ) gives a weight of 0.5

另一种是通过学习自动分配权重的方法:The other is by learning to assign weights automatically:

对每个领域cT收集一些有代表性的训练文本语料资料;Collect some representative training text corpora for each domain c T ;

对语料样本进行切词,生存领域cT的词库;Segment the words of the corpus samples, the lexicon of c T in the survival field;

计算领域cT中的词的权重,权重=TF×GIDF,其中TF为词在该领域cT所有语料中的词的总词频,GIDF为全局反向文档频率,GIDF=log(1+N/GDF),其中N为所有领域的所有文档的总数量,GDF为全局文档频率即为所有领域中包含该词的的所有文档的数量。Calculate the weight of the words in the field c T , weight=TF×GIDF, where TF is the total word frequency of words in all corpus of the field c T , GIDF is the global reverse document frequency, GIDF=log(1+N/ GDF), where N is the total number of all documents in all fields, and GDF is the global document frequency, which is the number of all documents containing the word in all fields.

设置各个水平的阈值,如T1,T2,...,Tn,T1>T2>...>TnSet thresholds for each level, such as T1, T2, ..., Tn, T1>T2>...>Tn

对领域cT词库中词根据其权重按上面阈值划分为多个档次的集合,Ti>总词频>Ti+1的为第个档次。For the field c, the words in the T lexicon are divided into sets of multiple grades according to their weights according to the above threshold, and the first grade is Ti>total word frequency>Ti+1.

对各个档次的词分别赋予一定的最终评分值,第一档赋予最高评分值,中间档赋予中间大小的评分值,第n档赋予最小评分值。A certain final score value is assigned to each grade of words, the first grade is assigned the highest score value, the middle grade is assigned a score value of an intermediate size, and the nth grade is assigned a minimum score value.

由词库中的词及其最终评分值组成领域cT向量。The domain c T vector is composed of words in the thesaurus and their final score values.

(2)给搜索请求的关键字赋予一定的权重,组成一个Query的向量,Query(q1,q2,...qn’)。(2) Assign a certain weight to the keyword of the search request to form a Query vector, Query(q1, q2,...qn').

Query的关键字的权重可以这样赋予:The weight of the keyword of Query can be given like this:

方法1:全部关键字赋予权重1;Method 1: assign weight 1 to all keywords;

方法2:排在最前面的关键字赋予最大权重(比如赋予权重1),排在中间的关键字赋予中间大小的权重(比如赋予0.5<权重<1),排在最后的关键字赋予最小权重(比如赋予权重0.5)。Method 2: The top keywords are given the greatest weight (for example, weight 1), the middle keywords are given middle weights (for example, 0.5<weight<1), and the last keywords are given the smallest weight (For example, assign a weight of 0.5).

(3)计算领域向量cT(t1,t2,...,tn)与查询向量qT(q1,q2,...,qn’)之间的Cousine相似度:(3) Calculate the Cousine similarity between domain vector c T (t1, t2, ..., tn) and query vector q T (q1, q2, ..., qn'):

SimSim (( qq TT (( qq 11 ,, qq 22 ,, .. .. .. ,, qnqn ,, )) ,, cc TT (( tt 11 ,, tt 22 ,, .. .. .. ,, tntn )) ))

== (( qq 11 &times;&times; tt 11 ++ qq 22 &times;&times; tt 22 ++ .. .. .. .. .. .. ++ qnqn &times;&times; tntn )) // (( qq 11 22 ++ qq 22 22 ++ .. .. .. ++ qnqn 22 &times;&times;

tt 11 22 ++ tt 22 22 ++ .. .. .. ++ tntn 22 ))

(4)g1(cT,qT)=sim(qT,cT);(4) g 1 (c T , q T ) = sim(q T , c T );

(5)从搜索历史数据中收集查询q(t)的所有用户点击历史文档UT={uT},其中uT为查询qT对应的某个用户点击搜索结果文档的向量,计算uT与cT的cousine相似度:(5) Collect all user click history documents U T ={u T } for query q(t) from the search history data, where u T is the vector of a certain user click search result document corresponding to query q T , and calculate u T Cousine similarity with c T :

simsim (( cc TT (( tt 11 ,, tt 22 ,, .. .. .. ,, tntn )) ,, uu TT (( uu 11 ,, uu 22 ,, .. .. .. ,, unun )) ))

== (( uu 11 &times;&times; tt 11 ++ uu 22 &times;&times; tt 22 ++ .. .. .. .. .. .. ++ unun &times;&times; tntn )) // (( uu 11 22 ++ uu 22 22 ++ .. .. .. ++ unun 22 &times;&times;

tt 11 22 ++ tt 22 22 ++ .. .. .. ++ tntn 22

(6) g 2 ( c T , q T ) = &Sigma; u T sim ( c T , u T ) | U T | (6) g 2 ( c T , q T ) = &Sigma; u T sim ( c T , u T ) | u T |

上下文相关的特征函数fk的选取:Selection of context-dependent feature function f k :

(1)直接关联(1) Direct association

设置查询序列对(qt-1,qt)的标记序列对为(ct-1,ct),本实施例用在给定查询序列对(qt-1,qt)前提下,标记序列对(ct-1,ct)出现的次数来计算f1(ct-1,ct,q)Set the tag sequence pair of the query sequence pair (q t-1 , q t ) to (c t-1 , c t ), this embodiment is used on the premise of a given query sequence pair (q t-1 , q t ), Count the occurrences of the sequence pair (c t-1 , c t ) to compute f 1 (c t-1 , c t , q)

ff 11 (( cc tt -- 11 ,, cc tt ,, qq )) == Oo (( cc tt -- 11 ,, cc tt )) Oo (( qq tt -- 11 ,, qq tt ))

其中O(ct-1,ct)为用在给定查询序列对(qt-1,qt)前提下,标记序列对(ct-1,ct)出现的次数。Where O(c t-1 , c t ) is the number of occurrences of the marker sequence pair (c t-1 , c t ) under the premise of a given query sequence pair (q t-1 , q t ).

O(qt-1,qt)为查询序列对(qt-1,qt)出现的总次数。O(q t-1 , q t ) is the total number of occurrences of the query sequence pair (q t-1 , q t ).

(2)利用分类目录树间接关联(2) Indirect association using classification tree

假设标记序列对(ct-1,ct)处于分类目录树的第n层,(ct-1,ct)的祖先节点对的集合为

Figure B2009102208107D0000441
1≤i≤n-1,本实施例用在给定查询序列对(qt-1,qt)前提下(ct-1,ct)的祖先节点对
Figure B2009102208107D0000442
出现的次数来计算f2(ct-1,ct,q):Assuming that the tag sequence pair (c t-1 , c t ) is at the nth level of the classification tree, the set of ancestor node pairs of (c t-1 , c t ) is
Figure B2009102208107D0000441
1≤i≤n-1, this embodiment uses the ancestor node pair of (c t-1 , c t ) under the premise of a given query sequence pair (q t-1 , q t )
Figure B2009102208107D0000442
occurrences to calculate f 2 (c t-1 , c t , q):

ff 22 (( cc tt -- 11 ,, cc tt ,, qq )) == &Sigma;&Sigma; ii == 11 nno -- 11 Oo (( aa cc tt -- 11 (( ii )) ,, aa cc tt (( ii )) )) Oo (( qq tt -- 11 ,, qq tt ))

其中,为在给定查询序列对(qt-1,qt)前提下(ct-1,ct)的祖先节点对

Figure B2009102208107D0000445
出现的次数,O(qt-1,qt)为查询序列对(qt-1,qt)出现的总次数。in, is the ancestor node pair of (c t-1 , c t ) given the query sequence pair (q t-1 , q t )
Figure B2009102208107D0000445
The number of occurrences, O(q t-1 , q t ) is the total number of occurrences of the query sequence pair (q t-1 , q t ).

本实施例中,搜索应用服务器通过提取用户的即时兴趣模型和长期兴趣模型,使得搜索服务器可以根据用户的即时兴趣模型和长期兴趣模型进行相应的搜索,从而用户提供个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search application server extracts the user's immediate interest model and long-term interest model, so that the search server can perform corresponding searches according to the user's immediate interest model and long-term interest model, so that the user can provide personalized information that meets the user's needs. And search result information with high matching accuracy.

图9为本发明搜索服务器实施例一的结构示意图,如图9所示,本实施例的搜索服务器可以包括:第一接收模块11、第一处理模块12以及第一搜索模块13。其中,第一接收模块11用于接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及即时兴趣模型和长期兴趣模型;第一处理模块12用于根据各成员引擎元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值;第一搜索模块13用于根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。FIG. 9 is a schematic structural diagram of Embodiment 1 of the search server of the present invention. As shown in FIG. 9 , the search server of this embodiment may include: a first receiving module 11 , a first processing module 12 and a first searching module 13 . Wherein, the first receiving module 11 is used to receive a search request, and the search request carries keyword information to be searched and an immediate interest model and a long-term interest model; The instant interest model and the long-term interest model calculate the correlation score value of the member engine; the first search module 13 is used to select one or more member engines to search the keyword information according to the correlation score value .

本实施例的搜索服务器,其实现原理与方法实施例一的实现原理相同,不再赘述。The realization principle of the search server in this embodiment is the same as that of the method embodiment 1, and will not be repeated here.

本实施例中,搜索服务器在请求成员引擎搜索关键词信息之前,可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎进行选择,从而能够选择获取与所需搜索的关键词信息以及即时兴趣模型和长期兴趣模型匹配较好的成员引擎对该关键词信息进行搜索,从而能够获取精度较高的搜索结果信息,进一步满足用户的搜索需求。In this embodiment, before the search server requests the member engine to search for keyword information, it can select the member engine according to the user's immediate interest model and long-term interest model, so as to be able to select and obtain the keyword information and the immediate interest model required for the search. A member engine that matches well with the long-term interest model searches for the keyword information, so as to obtain search result information with high precision, and further meet the user's search needs.

图10为本发明搜索服务器实施例二的结构示意图,如图10所示,本实施例的搜索服务器在图9所示的搜索服务器的基础上,进一步地,所述第一接收模块11接收的即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。FIG. 10 is a schematic structural diagram of the second embodiment of the search server of the present invention. As shown in FIG. 10 , the search server of this embodiment is based on the search server shown in FIG. The instant interest model is an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are composed of query sequences q(1),...,q(t) in the same search context session as the current query q(t) -1), the relevant data of q(t) is calculated, and the search context session is a preset period of time before the occurrence of the current query q(t) including the occurrence time of the current query q(t).

所述第一处理模块12可以包括:第一计算单元121和第一处理单元122,其中,第一计算单元121用于计算所述关键词信息与成员引擎的元索引信息之间的第一最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第一阈值且成员引擎的元索引信息与长期兴趣模型的相似度大于第二阈值的基础上,成员引擎的元索引信息与即时兴趣模型的第二最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第三阈值且成员引擎的元索引信息与即时兴趣模型的相似度大于第四阈值的基础上,成员引擎的元索引信息与长期兴趣模型的第三最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第五阈值的基础上,成员引擎的元索引信息与长期兴趣模型和即时兴趣模型的加权相加的结果向量的第四最大相似度;第一处理单元122用于根据第一最大相似度、第二最大相似度、第三最大相似度和第四最大相似度计算成员引擎的相似度评分值。The first processing module 12 may include: a first calculation unit 121 and a first processing unit 122, wherein the first calculation unit 121 is configured to calculate a first maximum value between the keyword information and the meta index information of the member engine Similarity: calculated on the basis that the similarity between the meta index information of the member engine and the keyword information is greater than the first threshold and the similarity between the meta index information of the member engine and the long-term interest model is greater than the second threshold, the meta index information of the member engine The second maximum similarity with the instant interest model; calculated on the basis that the similarity between the meta index information of the member engine and the keyword information is greater than the third threshold and the similarity between the meta index information of the member engine and the instant interest model is greater than the fourth threshold On the basis of the third maximum similarity between the meta index information of the member engine and the long-term interest model; calculated on the basis that the similarity between the meta index information of the member engine and the keyword information is greater than the fifth threshold, the meta index information of the member engine and the long-term interest model The fourth maximum similarity of the result vector of the weighted addition of the interest model and the instant interest model; the first processing unit 122 is used to Calculates the similarity score value of the membership engine.

本实施例的搜索服务器,其实现原理与方法实施例二的实现原理相同,不再赘述。The implementation principle of the search server in this embodiment is the same as that of the second method embodiment, and will not be repeated here.

本实施例中,搜索服务器在请求成员引擎搜索关键词信息之前,可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎进行选择,从而能够选择获取与所需搜索的关键词信息以及即时兴趣模型和长期兴趣模型匹配较好的成员引擎对该关键词信息进行搜索,从而能够获取精度较高的搜索结果信息,进一步满足用户的搜索需求。In this embodiment, before the search server requests the member engine to search for keyword information, it can select the member engine according to the user's immediate interest model and long-term interest model, so as to be able to select and obtain the keyword information and the immediate interest model required for the search. A member engine that matches well with the long-term interest model searches for the keyword information, so as to obtain search result information with high precision, and further meet the user's search needs.

图11为本发明搜索服务器实施例三的结构示意图,如图11所示,本实施例的搜索服务器可以包括:第二发送模块21、第二接收模块22以及第二处理模块23,其中,第二发送模块21用于向一个或多个成员引擎发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;第二接收模块22用于接收所述一个或多个成员引擎搜索所述关键词信息后,根据所述即时兴趣模型和长期兴趣模型获取并反馈的搜索结果信息以及与所述搜索结果信息对应的评分信息;第二处理模块23用于根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息通过所述第二发送模块发送给所述搜索应用服务器。FIG. 11 is a schematic structural diagram of the third embodiment of the search server of the present invention. As shown in FIG. 11, the search server of this embodiment may include: a second sending module 21, a second receiving module 22 and a second processing module 23, wherein The second sending module 21 is used to send a search request to one or more member engines, carrying the keyword information to be searched and the instant interest model and long-term interest model obtained by the search application server in the search request; the second receiving module 22 uses After receiving the keyword information searched by the one or more member engines, the search result information obtained and fed back according to the immediate interest model and the long-term interest model and the scoring information corresponding to the search result information; the second process Module 23 is used to re-score and sort the search result information according to the scoring information and related factor information, obtain the re-scoring and sorting search result information, and pass the re-scoring and sorting search result information through the The second sending module sends to the search application server.

进一步地,第二接收模块22还用于接收所述搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。第二接收模块22接收的即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。Further, the second receiving module 22 is also configured to receive the search request sent by the search application server, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server. The instant interest model received by the second receiving module 22 is an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are composed of the query sequence q(1) in the same search context session as the current query q(t), ..., q(t-1), the relevant data of q(t) are calculated, and the search context session is a preset period including the occurrence time of the current query q(t) before the occurrence of the current query q(t) time.

本实施例的搜索服务器,其实现原理与方法实施例三和五的实现原理相同,不再赘述。The implementation principle of the search server in this embodiment is the same as that of method embodiments 3 and 5, and will not be repeated here.

本实施例中,搜索服务器可以将搜索应用服务器提取的即时兴趣模型和长期兴趣模型发送给成员引擎,从而使得成员引擎在获取搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化的评分处理,从而获取各搜索结果信息相应的评分信息。当搜索服务器接收到成员引擎反馈的搜索结果信息以及相应的评分信息以后,还可以结合其它相关因素对搜索结果信息进行重新评分排序,从而可以获取个性化的、满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can send the immediate interest model and long-term interest model extracted by the search application server to the member engine, so that after the member engine obtains the search result information, it can use the user's immediate interest model and long-term interest model to the member engine. The search result information is subjected to personalized scoring processing, so as to obtain the corresponding scoring information of each search result information. After the search server receives the search result information fed back by the member engine and the corresponding scoring information, it can also re-score and sort the search result information in combination with other relevant factors, so as to obtain personalized, satisfying user needs and high matching accuracy. Search result information.

图12为本发明搜索服务器实施例四的结构示意图,如图12所示,本实施例的搜索服务器可以包括:第三接收模块31、第三处理模块32以及第三发送模块33,其中,第三接收模块31用于接收搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;并接收成员引擎根据所述关键词信息搜索获取的搜索结果信息;第三处理模块32用于根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;第三发送模块33用于将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述搜索应用服务器。Fig. 12 is a schematic structural diagram of Embodiment 4 of the search server of the present invention. As shown in Fig. 12, the search server of this embodiment may include: a third receiving module 31, a third processing module 32 and a third sending module 33, wherein The third receiving module 31 is used to receive the search request sent by the search application server, which carries the keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server in the search request; The search result information obtained by word information search; the third processing module 32 is used to perform scoring and sorting processing on the search result information according to the instant interest model and the long-term interest model; the third sending module 33 is used to process the scoring sorting The search result information and corresponding scoring information are sent to the search application server.

本实施例的搜索服务器,其实现原理与方法实施例六的实现原理相同,不再赘述。The realization principle of the search server in this embodiment is the same as the realization principle of the sixth method embodiment, and will not be repeated here.

本实施例中,搜索服务器可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎反馈的搜索结果信息进行评分排序,从而用户提供个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can rank and rank the search result information fed back by the member engine according to the user's immediate interest model and long-term interest model, so that the user can provide personalized search result information that meets the user's needs and has high matching accuracy.

图13为本发明搜索服务器实施例五的结构示意图,如图13所示,本实施例的搜索服务器在图12所示的搜索服务器的基础上,进一步地,第三处理模块32可以包括:第三计算单元321和第三处理单元322,其中,第三计算单元321用于计算所述即时兴趣模型与所述搜索结果信息的第一相似度;计算所述长期兴趣模型与所述搜索结果信息的第二相似度;第三处理单元322用于根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。FIG. 13 is a schematic structural diagram of Embodiment 5 of the search server of the present invention. As shown in FIG. 13 , the search server of this embodiment is based on the search server shown in FIG. 12 , and further, the third processing module 32 may include: Three calculation units 321 and a third processing unit 322, wherein the third calculation unit 321 is used to calculate the first similarity between the instant interest model and the search result information; calculate the long-term interest model and the search result information The second similarity degree; the third processing unit 322 is configured to obtain a score value according to the first similarity degree and the second similarity degree, and sort the search result information according to the score value.

本实施例的搜索服务器,其实现原理与方法实施例七的实现原理相同,不再赘述。The implementation principle of the search server in this embodiment is the same as that in the seventh embodiment of the method, and will not be repeated here.

本实施例中,搜索服务器可以根据用户的即时兴趣模型和长期兴趣模型对成员引擎反馈的搜索结果信息进行评分排序,从而用户提供个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search server can rank and rank the search result information fed back by the member engine according to the user's immediate interest model and long-term interest model, so that the user can provide personalized search result information that meets the user's needs and has high matching accuracy.

图14为本发明成员引擎设备实施例一的结构示意图,如图14所示,本实施例的成员引擎设备可以包括:第四接收模块41、第四处理模块42以及第四发送模块43,其中,第四接收模块41用于接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及即时兴趣模型和长期兴趣模型;第四处理模块42用于根据所述搜索请求对所述关键词信息进行搜索获取搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;第四发送模块43用于返回评分排序处理后的搜索结果信息。FIG. 14 is a schematic structural diagram of Embodiment 1 of the member engine device of the present invention. As shown in FIG. 14 , the member engine device of this embodiment may include: a fourth receiving module 41, a fourth processing module 42, and a fourth sending module 43, wherein , the fourth receiving module 41 is used to receive a search request, and the search request carries keyword information to be searched and an immediate interest model and a long-term interest model; the fourth processing module 42 is used to search the keyword according to the search request Word information is searched to obtain search result information, and the search result information is scored and sorted according to the immediate interest model and the long-term interest model; the fourth sending module 43 is used to return the scored and sorted search result information.

本实施例的成员引擎设备,其实现原理与方法实施例四的实现原理相同,不再赘述。The implementation principle of the member engine device in this embodiment is the same as that in the fourth embodiment of the method, and will not be repeated here.

本实施例中,成员引擎设备可以接收搜索服务器发送的即时兴趣模型和长期兴趣模型,从而在获取与关键词信息对应的搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化的评分处理,获取各搜索结果信息相应的评分信息。从而可以获取个性化的、满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the member engine device can receive the immediate interest model and the long-term interest model sent by the search server, so that after obtaining the search result information corresponding to the keyword information, the search engine can be based on the user's immediate interest model and long-term interest model. The result information is subjected to personalized scoring processing, and the corresponding scoring information of each search result information is obtained. In this way, personalized search result information that satisfies user needs and has high matching accuracy can be obtained.

图15为本发明成员引擎设备实施例二的结构示意图,如图15所示,本实施例的成员引擎设备在图14所示的成员引擎设备的基础上,进一步地,第四处理模块42可以包括:第四计算单元421和第四处理单元422,其中,第四计算单元421用于计算所述即时兴趣模型与所述搜索结果信息的第一相似度;计算所述长期兴趣模型与所述搜索结果信息的第二相似度;第四处理单元422用于根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。FIG. 15 is a schematic structural diagram of Embodiment 2 of the member engine device of the present invention. As shown in FIG. 15 , the member engine device of this embodiment is based on the member engine device shown in FIG. 14 , further, the fourth processing module 42 can Including: a fourth calculation unit 421 and a fourth processing unit 422, wherein the fourth calculation unit 421 is used to calculate the first similarity between the instant interest model and the search result information; calculate the long-term interest model and the The second similarity degree of the search result information; the fourth processing unit 422 is configured to obtain a score value according to the first similarity degree and the second similarity degree, and sort the search result information according to the score value.

本实施例的成员引擎设备,其实现原理与方法实施例四和五的实现原理相同,不再赘述。The implementation principle of the member engine device in this embodiment is the same as that of method embodiments 4 and 5, and will not be repeated here.

本实施例中,成员引擎设备可以接收搜索服务器发送的即时兴趣模型和长期兴趣模型,从而在获取与关键词信息对应的搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化的评分处理,获取各搜索结果信息相应的评分信息。从而可以获取个性化的、满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the member engine device can receive the immediate interest model and the long-term interest model sent by the search server, so that after obtaining the search result information corresponding to the keyword information, the search engine can be based on the user's immediate interest model and long-term interest model. The result information is subjected to personalized scoring processing, and the corresponding scoring information of each search result information is obtained. In this way, personalized search result information that satisfies user needs and has high matching accuracy can be obtained.

图16为本发明搜索应用服务器实施例的结构示意图,如图16所示,本实施例的搜索应用服务器可以包括:第五接收模块51、第五处理模块52以及第五发送模块53,其中,第五接收模块51用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;第五处理模块52用于从用户数据库中提取即时兴趣模型和长期兴趣模型;第五发送模块53用于向搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型,以使所述搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。FIG. 16 is a schematic structural diagram of an embodiment of the search application server of the present invention. As shown in FIG. 16, the search application server of this embodiment may include: a fifth receiving module 51, a fifth processing module 52, and a fifth sending module 53, wherein, The fifth receiving module 51 is used to receive the search request message sent by the search client, carrying keyword information in the search request message; the fifth processing module 52 is used to extract the immediate interest model and the long-term interest model from the user database; the fifth The sending module 53 is used to send a search request to the search server, and the search request carries keyword information and the immediate interest model and the long-term interest model, so that the search server searches the search server according to the immediate interest model and the long-term interest model. Search for the above keyword information.

进一步地,第五处理模块52具体用于应用条件随机场模型计算在给定与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),....q(t-1),q(t)的条件下,当前查询q(t)的输出类型的条件概率,将该条件概率值作为即时兴趣模型与该输出类型对应的兴趣维度的评分值。Further, the fifth processing module 52 is specifically used to apply the conditional random field model to calculate the query sequence q(1), ... q(t-1) in a given session in the same search context as the current query q(t). , under the condition of q(t), currently query the conditional probability of the output type of q(t), and use the conditional probability value as the score value of the interest dimension corresponding to the output type of the instant interest model.

本实施例的搜索应用服务器,其实现原理与方法实施例八的实现原理相同,不再赘述。The implementation principle of the search application server in this embodiment is the same as that in the eighth method embodiment, and will not be repeated here.

本实施例中,搜索应用服务器通过提取用户的即时兴趣模型和长期兴趣模型,使得搜索服务器可以根据用户的即时兴趣模型和长期兴趣模型进行相应的搜索,从而用户提供个性化的,满足用户需求的以及匹配精度高的搜索结果信息。In this embodiment, the search application server extracts the user's immediate interest model and long-term interest model, so that the search server can perform corresponding searches according to the user's immediate interest model and long-term interest model, so that the user can provide personalized information that meets the user's needs. And search result information with high matching accuracy.

图17为本发明移动搜索系统实施例一的结构示意图,如图17所示,本实施例的系统可以包括:第一搜索应用服务器1、第一搜索服务器2以及第一成员引擎设备3,其中,第一搜索应用服务器1用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第一搜索服务器2发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;第一搜索服务器2用于接收所述第一搜索应用服务器1发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及第一搜索应用服务器1获取的即时兴趣模型和长期兴趣模型;根据关键词信息、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值,根据所述相关度评分值从第一成员引擎设备3中选择一个或多个成员引擎对所述关键词信息进行搜索;第一成员引擎设备3用于接收所述第一搜索服务器2发送的搜索请求,对所述关键词信息进行搜索,并将搜索结果信息发送给所述第一搜索服务器2,以使所述第一搜索服务器2通过所述第一搜索应用服务器1将所述搜索结果信息反馈给所述搜索客户端。FIG. 17 is a schematic structural diagram of Embodiment 1 of the mobile search system of the present invention. As shown in FIG. 17, the system of this embodiment may include: a first search application server 1, a first search server 2, and a first member engine device 3, wherein , the first search application server 1 is used to receive the search request message sent by the search client, the search request message carries keyword information; extract the immediate interest model and the long-term interest model from the user database; send to the first search server 2 A search request, carrying keyword information and the immediate interest model and long-term interest model in the search request; the first search server 2 is configured to receive the search request sent by the first search application server 1, carrying in the search request The keyword information to be searched and the immediate interest model and long-term interest model obtained by the first search application server 1; according to the keyword information, the meta index information of each member engine and the immediate interest model and long-term interest model, calculate the According to the correlation score value of the member engine, one or more member engines are selected from the first member engine device 3 to search for the keyword information according to the correlation score value; the first member engine device 3 is used to receive the The search request sent by the first search server 2 searches the keyword information, and sends the search result information to the first search server 2, so that the first search server 2 can use the first search application Server 1 feeds back the search result information to the search client.

本实施例的移动搜索系统中,第一搜索服务器在请求第一成员引擎设备搜索关键词信息之前,可以根据第一搜索应用服务器发送的用户的即时兴趣模型和长期兴趣模型对第一成员引擎设备进行选择,从而能够选择获取与所需搜索的关键词信息匹配以及即时兴趣模型和长期兴趣模型较好的第一成员引擎设备中的成员引擎对该关键词信息进行搜索,从而能够获取精度较高的搜索结果信息,进一步满足用户的搜索需求。In the mobile search system of this embodiment, before the first search server requests the first member engine device to search for keyword information, it can search the first member engine device according to the user's immediate interest model and long-term interest model sent by the first search application server. Select, so that the member engine in the first member engine device that matches the keyword information that needs to be searched and has a better immediate interest model and long-term interest model can be selected to search for the keyword information, so that the keyword information can be obtained with higher accuracy search result information to further meet the user's search needs.

图18为本发明移动搜索系统实施例二的结构示意图,如图18所示,本实施例的系统可以包括:第二搜索应用服务器4、第二搜索服务器5以及第二成员引擎设备6,其中,第二搜索应用服务器4用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第二搜索服务器5发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;第二搜索服务器5用于向第二成员引擎设备6发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及第二搜索应用服务器4获取的即时兴趣模型和长期兴趣模型;接收所述第二成员引擎设备6反馈的搜索结果信息以及与所述搜索结果信息对应的评分信息;根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息发送给所述第二搜索应用服务器4;第二成员引擎设备6用于接收所述第二搜索服务器5发送的搜索请求,对所述关键词信息进行搜索,根据所述即时兴趣模型和长期兴趣模型获取搜索结果信息以及与所述搜索结果信息对应的评分信息,并将所述搜索结果信息以及评分信息发送给所述第二搜索服务器5。Fig. 18 is a schematic structural diagram of Embodiment 2 of the mobile search system of the present invention. As shown in Fig. 18, the system of this embodiment may include: a second search application server 4, a second search server 5, and a second member engine device 6, wherein , the second search application server 4 is used to receive the search request message sent by the search client, carrying keyword information in the search request message; extract the instant interest model and the long-term interest model from the user database; send to the second search server 5 A search request, carrying keyword information and the instant interest model and long-term interest model in the search request; the second search server 5 is used to send a search request to the second member engine device 6, carrying the required search request in the search request keyword information and the immediate interest model and long-term interest model acquired by the second search application server 4; receive the search result information fed back by the second member engine device 6 and the scoring information corresponding to the search result information; according to the Score information and related factor information re-score and sort the search result information, obtain the re-score and sort search result information, and send the re-score and sort search result information to the second search application server 4; The second member engine device 6 is used to receive the search request sent by the second search server 5, search for the keyword information, obtain search result information according to the instant interest model and the long-term interest model and match the search result information corresponding to the scoring information, and send the search result information and scoring information to the second search server 5 .

本实施例的移动搜索系统中,第二搜索服务器可以将第二搜索应用服务器提取的即时兴趣模型和长期兴趣模型发送给第二成员引擎设备,从而使得第二成员引擎设备在获取搜索结果信息后,可以根据用户的即时兴趣模型和长期兴趣模型对该搜索结果信息进行个性化评分处理,从而获取各搜索结果信息相应的评分信息。当第二搜索服务器接收到第二成员引擎设备反馈的搜索结果信息以及相应的评分信息以后,还可以结合其它相关因素对搜索结果信息进行重新评分排序,从而可以获取个性化的、满足用户需求的以及匹配精度高的搜索结果信息。In the mobile search system of this embodiment, the second search server can send the instant interest model and long-term interest model extracted by the second search application server to the second member engine device, so that the second member engine device obtains the search result information According to the user's immediate interest model and long-term interest model, the search result information can be personalized to score, so as to obtain the corresponding score information of each search result information. After the second search server receives the search result information fed back by the second member engine device and the corresponding scoring information, it can also re-rank and rank the search result information in combination with other relevant factors, so as to obtain personalized information that meets user needs. And search result information with high matching accuracy.

图19为本发明移动搜索系统实施例三的结构示意图,如图19所示,本实施例的系统可以包括:第三搜索应用服务器7、第三搜索服务器8以及第三成员引擎设备9,其中,第三搜索应用服务器7用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第三搜索服务器8发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;第三搜索服务器8用于接收第三搜索应用服务器7发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;接收第三成员引擎设备9根据所述关键词信息搜索获取的搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;并将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述第三搜索应用服务器7;第三成员引擎设备9用于接收所述第三搜索服务器8发送的搜索请求,对所述关键词信息进行搜索,获取所述搜索结果信息。FIG. 19 is a schematic structural diagram of Embodiment 3 of the mobile search system of the present invention. As shown in FIG. 19 , the system of this embodiment may include: a third search application server 7, a third search server 8, and a third member engine device 9, wherein , the third search application server 7 is used to receive the search request message sent by the search client, carrying keyword information in the search request message; extract the instant interest model and long-term interest model from the user database; send to the third search server 8 A search request, carrying keyword information and the instant interest model and long-term interest model in the search request; the third search server 8 is used to receive the search request sent by the third search application server 7, carrying the required The keyword information of the search and the immediate interest model and long-term interest model obtained by the search application server; receiving the search result information obtained by the third member engine device 9 according to the keyword information search, and according to the immediate interest model and the long-term interest model Perform scoring and sorting processing on the search result information; and send the search result information after the scoring and sorting process and corresponding scoring information to the third search application server 7; the third member engine device 9 is used to receive the third Search the search request sent by the server 8 to search the keyword information to obtain the search result information.

本实施例的移动搜索系统,第三搜索服务器可以根据第三搜索应用服务器提取的用户的即时兴趣模型和长期兴趣模型对第三成员引擎设备反馈的搜索结果信息进行评分排序,从而用户提供个性化的、满足用户需求的以及匹配精度高的搜索结果信息。In the mobile search system of this embodiment, the third search server can score and sort the search result information fed back by the third member engine device according to the user's immediate interest model and long-term interest model extracted by the third search application server, so that the user provides personalized Search result information that meets user needs and has high matching accuracy.

本发明的各个装置、服务器和系统实施例中所提供各个单元之间的交互及相关信息均可以参考前述各个方法实施例提供的相关流程,具体功能和处理流程请参见前述各个实施例,此处不再赘述。For the interaction and related information between the various units provided in the various device, server and system embodiments of the present invention, you can refer to the relevant processes provided by the foregoing method embodiments. For specific functions and processing procedures, please refer to the foregoing embodiments. Here No longer.

通过以上的实施方式的描述,所属领域的技术人员可以清楚地了解到本发明可借助软件加必需的通用硬件的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本发明的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在可读取的存储介质中,如计算机的软盘,硬盘或光盘等,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本发明各个实施例所述的方法。Through the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be realized by means of software plus necessary general-purpose hardware, and of course also by hardware, but in many cases the former is a better embodiment . Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a readable storage medium, such as a floppy disk of a computer , a hard disk or an optical disk, etc., including several instructions for enabling a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

最后应说明的是:以上实施例仅用以说明本发明的技术方案,而非对其限制;尽管参照前述实施例对本发明进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (33)

1.一种移动搜索方法,其特征在于,包括:1. A mobile search method, characterized in that, comprising: 接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型;Receiving a search request, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server; 根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值;According to the search request, the meta index information of each member engine, and the instant interest model and long-term interest model, calculate the relevance score value of the member engine; 根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。Selecting one or more member engines to search the keyword information according to the correlation score value. 2.根据权利要求1所述的移动搜索方法,其特征在于,所述即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与所述搜索请求对应的当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。2. The mobile search method according to claim 1, wherein the instant interest model is an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by the current interest model corresponding to the search request. Query q(t) is calculated from the relevant data of the query sequence q(1),...,q(t-1), q(t) in the same search context session, where the search context session is the current query q(t ) is a preset period of time including the occurrence time of the current query q(t). 3.根据权利要求1所述的移动搜索方法,其特征在于,所述长期兴趣模型为N个维度的评分值所组成的长期兴趣模型向量,各个维度的评分值由用户搜索历史数据和静态档案profile计算得到。3. The mobile search method according to claim 1, wherein the long-term interest model is a long-term interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by user search history data and static archives profile is calculated. 4.根据权利要求1~3中任一权利要求所述的移动搜索方法,其特征在于,所述根据各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值,包括:4. The mobile search method according to any one of claims 1 to 3, characterized in that, according to the meta index information of each member engine and the immediate interest model and long-term interest model, the member engine is calculated Relevance score values for , including: 计算所述关键词信息与成员引擎的元索引信息之间的第一最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第一阈值且成员引擎的元索引信息与长期兴趣模型的相似度大于第二阈值的基础上,成员引擎的元索引信息与即时兴趣模型的第二最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第三阈值且成员引擎的元索引信息与即时兴趣模型的相似度大于第四阈值的基础上,成员引擎的元索引信息与长期兴趣模型的第三最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第五阈值的基础上,成员引擎的元索引信息与长期兴趣模型和即时兴趣模型的加权相加的结果向量的第四最大相似度;Calculate the first maximum similarity between the keyword information and the meta index information of the member engine; the calculated similarity between the meta index information of the member engine and the keyword information is greater than the first threshold and the meta index information of the member engine and the long-term On the basis that the similarity of the interest model is greater than the second threshold, the second maximum similarity between the meta index information of the member engine and the instant interest model; On the basis that the similarity between the meta index information of the member engine and the instant interest model is greater than the fourth threshold, the third maximum similarity between the meta index information of the member engine and the long-term interest model; calculated in the meta index information and keyword information of the member engine On the basis that the similarity of is greater than the fifth threshold, the meta index information of the member engine and the result vector of the weighted addition of the long-term interest model and the instant interest model have the fourth maximum similarity; 根据第一最大相似度、第二最大相似度、第三最大相似度和第四最大相似度计算成员引擎的相似度评分值。Calculate the similarity score value of the member engine according to the first maximum similarity, the second maximum similarity, the third maximum similarity and the fourth maximum similarity. 5.根据权利要求4所述的移动搜索方法,其特征在于,所述根据第一最大相似度、第二最大相似度、第三最大相似度和第四最大相似度计算成员引擎的相似度评分值,包括:5. The mobile search method according to claim 4, wherein the similarity score of the member engine is calculated according to the first maximum similarity, the second maximum similarity, the third maximum similarity and the fourth maximum similarity values, including: 对第一最大相似度、第二最大相似度、第三最大相似度和第四最大相似度加权相加、或相乘、或取最大值,获取相似度评分值。The first maximum similarity, the second maximum similarity, the third maximum similarity and the fourth maximum similarity are weighted, added, or multiplied, or the maximum value is obtained to obtain a similarity score. 6.一种移动搜索方法,其特征在于,包括:6. A mobile search method, comprising: 向一个或多个成员引擎发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;Send a search request to one or more member engines, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server; 接收所述一个或多个成员引擎根据所述关键词信息、所述即时兴趣模型和长期兴趣模型获取的搜索结果信息以及与所述搜索结果信息对应的评分信息;receiving search result information obtained by the one or more member engines according to the keyword information, the immediate interest model and the long-term interest model, and scoring information corresponding to the search result information; 根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息发送给所述搜索应用服务器。Re-scoring and sorting the search result information according to the scoring information and relevant factor information, obtaining the re-scoring and sorting search result information, and sending the re-scoring and sorting search result information to the search application server. 7.根据权利要求6所述的移动搜索方法,其特征在于,还包括:7. The mobile search method according to claim 6, further comprising: 接收所述搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。Receive the search request sent by the search application server, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server. 8.根据权利要求7所述的移动搜索方法,其特征在于,所述即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。8. mobile search method according to claim 7, is characterized in that, described immediate interest model is the instant interest model vector that the scoring value of N dimension forms, and the scoring value of each dimension is by and current inquiry q (t) is in The relevant data of the query sequence q(1), ..., q(t-1), q(t) of the same search context session is calculated, and the search context session includes the current query before the occurrence of the current query q(t) A predetermined period of time within the occurrence time of q(t). 9.根据权利要求7所述的移动搜索方法,其特征在于,所述长期兴趣模型为N个维度的评分值所组成的长期兴趣模型向量,各个维度的评分值由用户搜索历史数据和静态档案profile计算得到。9. The mobile search method according to claim 7, wherein the long-term interest model is a long-term interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by user search history data and static archives profile is calculated. 10.根据权利要求6~9中任一权利要求所述的移动搜索方法,其特征在于,所述相关因素信息包括:成员引擎级别信息和/或价格信息,所述根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,包括:10. The mobile search method according to any one of claims 6-9, wherein the relevant factor information includes: member engine level information and/or price information, and the The factor information re-scores and sorts the search result information, including: 根据所述评分信息、成员引擎级别信息和/或价格信息,计算所述搜索结果的综合评分值,并根据所述综合评分值对所述搜索结果信息进行排序处理。According to the score information, member engine level information and/or price information, the comprehensive score value of the search result is calculated, and the search result information is sorted according to the comprehensive score value. 11.一种移动搜索方法,其特征在于,包括:11. A mobile search method, comprising: 接收搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;Receive the search request sent by the search application server, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server; 接收成员引擎根据所述关键词信息搜索获取的搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;receiving the search result information obtained by the member engine according to the keyword information search, and performing scoring and sorting processing on the search result information according to the instant interest model and the long-term interest model; 将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述搜索应用服务器。Send the search result information after scoring and sorting processing and the corresponding scoring information to the search application server. 12.根据权利要求11所述的移动搜索方法,其特征在于,所述即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与所述搜索请求对应的当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。12. The mobile search method according to claim 11, wherein the instant interest model is an instant interest model vector composed of score values of N dimensions, and the score values of each dimension are determined by the current interest model corresponding to the search request. Query q(t) is calculated from the relevant data of the query sequence q(1),...,q(t-1), q(t) in the same search context session, where the search context session is the current query q(t ) is a preset period of time including the occurrence time of the current query q(t). 13.根据权利要求11所述的移动搜索方法,其特征在于,所述长期兴趣模型为N个维度的评分值所组成的长期兴趣模型向量,各个维度的评分值由用户搜索历史数据和静态档案profile计算得到。13. The mobile search method according to claim 11, wherein the long-term interest model is a long-term interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by user search history data and static archives profile is calculated. 14.根据权利要求11~13中任一权利要求所述的移动搜索方法,其特征在于,所述根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理,包括:14. The mobile search method according to any one of claims 11-13, wherein the scoring and sorting of the search result information according to the instant interest model and the long-term interest model includes: 计算所述即时兴趣模型与所述搜索结果信息的第一相似度;calculating a first similarity between the instant interest model and the search result information; 计算所述长期兴趣模型与所述搜索结果信息的第二相似度;calculating a second similarity between the long-term interest model and the search result information; 根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。A score value is acquired according to the first similarity degree and the second similarity degree, and the search result information is sorted according to the score value. 15.一种移动搜索方法,其特征在于,包括:15. A mobile search method, comprising: 接收搜索请求,所述搜索请求中携带即时兴趣模型和长期兴趣模型;Receiving a search request, the search request carries an immediate interest model and a long-term interest model; 根据所述搜索请求进行搜索获取搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;performing a search according to the search request to obtain search result information, and performing scoring and ranking processing on the search result information according to the immediate interest model and the long-term interest model; 返回评分排序处理后的搜索结果信息。Returns the search result information after scoring and sorting. 16.根据权利要求15所述的移动搜索方法,其特征在于,所述根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理,包括:16. The mobile search method according to claim 15, wherein the scoring and sorting of the search result information according to the instant interest model and the long-term interest model comprises: 计算所述即时兴趣模型与所述搜索结果信息的第一相似度;calculating a first similarity between the instant interest model and the search result information; 计算所述长期兴趣模型与所述搜索结果信息的第二相似度;calculating a second similarity between the long-term interest model and the search result information; 根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。A score value is acquired according to the first similarity degree and the second similarity degree, and the search result information is sorted according to the score value. 17.一种移动搜索方法,其特征在于,包括:17. A mobile search method, comprising: 接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;receiving a search request message sent by a search client, wherein the search request message carries keyword information; 从用户数据库中提取即时兴趣模型和长期兴趣模型;Extract immediate and long-term interest models from the user database; 向搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型,以使所述搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。Sending a search request to a search server, the search request carrying keyword information and the immediate interest model and long-term interest model, so that the search server can perform a search on the keyword information according to the immediate interest model and the long-term interest model search. 18.根据权利要求17所述的移动搜索方法,其特征在于,所述从用户数据库中提取即时兴趣模型,包括:18. The mobile search method according to claim 17, wherein said extracting the instant interest model from the user database comprises: 应用条件随机场模型计算在给定与所述搜索请求对应的当前查询q(t)处于同一搜索上下文会话的查询序列q(1),....q(t-1),q(t)的条件下,当前查询q(t)的输出类型的条件概率,将该条件概率值作为即时兴趣模型与该输出类型对应的兴趣维度的评分值。Applying the conditional random field model to calculate the query sequence q(1), ... q(t-1), q(t) given that the current query q(t) corresponding to the search request is in the same search context session Under the condition of , the conditional probability of the output type of the current query q(t), the conditional probability value is used as the score value of the interest dimension corresponding to the output type of the instant interest model. 19.一种搜索服务器,其特征在于,包括:19. A search server, comprising: 第一接收模块,用于接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述搜索应用服务器获取的即时兴趣模型和长期兴趣模型;The first receiving module is configured to receive a search request, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server; 第一处理模块,用于根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值;A first processing module, configured to calculate the relevance score value of the member engine according to the search request, the meta index information of each member engine, and the immediate interest model and long-term interest model; 第一搜索模块,用于根据所述相关度评分值选择一个或多个成员引擎对所述关键词信息进行搜索。A first search module, configured to select one or more member engines to search the keyword information according to the correlation score. 20.根据权利要求19所述的搜索服务器,其特征在于,所述第一接收模块接收的即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与所述搜索请求对应的当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。20. The search server according to claim 19, wherein the instant interest model received by the first receiving module is an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by the The current query q(t) corresponding to the search request is calculated from the relevant data of the query sequence q(1),...,q(t-1), q(t) in the same search context session, and the search context session is A period of preset time including the occurrence time of the current query q(t) before the occurrence of the current query q(t). 21.根据权利要求19或20所述的搜索服务器,其特征在于,所述第一处理模块包括:21. The search server according to claim 19 or 20, wherein the first processing module comprises: 第一计算单元,用于计算所述关键词信息与成员引擎的元索引信息之间的第一最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第一阈值且成员引擎的元索引信息与长期兴趣模型的相似度大于第二阈值的基础上,成员引擎的元索引信息与即时兴趣模型的第二最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第三阈值且成员引擎的元索引信息与即时兴趣模型的相似度大于第四阈值的基础上,成员引擎的元索引信息与长期兴趣模型的第三最大相似度;计算在成员引擎的元索引信息与关键词信息的相似度大于第五阈值的基础上,成员引擎的元索引信息与长期兴趣模型和即时兴趣模型的加权相加的结果向量的第四最大相似度;The first calculation unit is used to calculate the first maximum similarity between the keyword information and the meta index information of the member engine; the calculated similarity between the meta index information of the member engine and the keyword information is greater than the first threshold and the member On the basis that the similarity between the meta index information of the engine and the long-term interest model is greater than the second threshold, the second maximum similarity between the meta index information of the member engine and the instant interest model; On the basis that the similarity is greater than the third threshold and the similarity between the meta index information of the member engine and the instant interest model is greater than the fourth threshold, the third maximum similarity between the meta index information of the member engine and the long-term interest model; On the basis that the similarity between the meta index information and the keyword information is greater than the fifth threshold, the fourth maximum similarity between the meta index information of the member engine and the result vector of the weighted addition of the long-term interest model and the instant interest model; 第一处理单元,用于根据第一最大相似度、第二最大相似度、第三最大相似度和第四最大相似度计算成员引擎的相似度评分值。The first processing unit is configured to calculate the similarity score value of the member engine according to the first maximum similarity, the second maximum similarity, the third maximum similarity and the fourth maximum similarity. 22.一种搜索服务器,其特征在于,包括:22. A search server, comprising: 第二发送模块,用于向一个或多个成员引擎发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;The second sending module is used to send a search request to one or more member engines, and the search request carries the keyword information to be searched and the immediate interest model and long-term interest model acquired by the search application server; 第二接收模块,用于接收所述一个或多个成员引擎根据所述关键词信息、所述即时兴趣模型和长期兴趣模型获取的搜索结果信息以及与所述搜索结果信息对应的评分信息;A second receiving module, configured to receive search result information obtained by the one or more member engines according to the keyword information, the immediate interest model and the long-term interest model, and scoring information corresponding to the search result information; 第二处理模块,用于根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息通过所述第二发送模块发送给所述搜索应用服务器。The second processing module is used to re-score and sort the search result information according to the scoring information and related factor information, obtain the re-scoring and sorting search result information, and pass the re-scoring and sorting search result information through The second sending module sends to the search application server. 23.根据权利要求22所述的搜索服务器,其特征在于,所述第二接收模块还用于接收所述搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型。23. The search server according to claim 22, wherein the second receiving module is further configured to receive a search request sent by the search application server, the search request carries keyword information to be searched and Search the immediate interest model and long-term interest model acquired by the application server. 24.根据权利要求23所述的搜索服务器,其特征在于,所述第二接收模块接收的即时兴趣模型为N个维度的评分值组成的即时兴趣模型向量,各个维度的评分值由与所述接收的搜索请求对应的当前查询q(t)处于同一搜索上下文会话的查询序列q(1),...,q(t-1),q(t)的相关数据计算得到,所述搜索上下文会话为当前查询q(t)发生之前包括当前查询q(t)发生时间在内的一段预设时间。24. The search server according to claim 23, wherein the instant interest model received by the second receiving module is an instant interest model vector composed of scoring values of N dimensions, and the scoring values of each dimension are determined by the The received search request corresponding to the current query q(t) is in the same search context session query sequence q(1), ..., q(t-1), q(t) related data is calculated, the search context A session is a preset period of time including the occurrence time of the current query q(t) before the occurrence of the current query q(t). 25.一种搜索服务器,其特征在于,包括:25. A search server, comprising: 第三接收模块,用于接收搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;并接收成员引擎根据所述关键词信息搜索获取的搜索结果信息;The third receiving module is used to receive the search request sent by the search application server, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server; Search result information obtained by keyword information search; 第三处理模块,用于根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;A third processing module, configured to perform scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model; 第三发送模块,用于将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述搜索应用服务器。The third sending module is configured to send the search result information after scoring and sorting processing and corresponding scoring information to the search application server. 26.根据权利要求25所述的搜索服务器,其特征在于,所述第三处理模块包括:26. The search server according to claim 25, wherein the third processing module comprises: 第三计算单元,用于计算所述即时兴趣模型与所述搜索结果信息的第一相似度;计算所述长期兴趣模型与所述搜索结果信息的第二相似度;A third calculation unit, configured to calculate a first similarity between the instant interest model and the search result information; calculate a second similarity between the long-term interest model and the search result information; 第三处理单元,用于根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。The third processing unit is configured to acquire a score value according to the first similarity degree and the second similarity degree, and sort the search result information according to the score value. 27.一种成员引擎设备,其特征在于,包括:27. A member engine device, characterized in that it comprises: 第四接收模块,用于接收搜索请求,所述搜索请求中携带所需搜索的关键词信息以及即时兴趣模型和长期兴趣模型;The fourth receiving module is used to receive a search request, and the search request carries keyword information to be searched and an immediate interest model and a long-term interest model; 第四处理模块,用于根据所述搜索请求对所述关键词信息进行搜索获取搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;A fourth processing module, configured to search the keyword information according to the search request to obtain search result information, and perform scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model; 第四发送模块,用于返回评分排序处理后的搜索结果信息。The fourth sending module is used to return the search result information after scoring and sorting. 28.根据权利要求27所述的搜索服务器,其特征在于,所述第四处理模块包括:28. The search server according to claim 27, wherein the fourth processing module comprises: 第四计算单元,用于计算所述即时兴趣模型与所述搜索结果信息的第一相似度;计算所述长期兴趣模型与所述搜索结果信息的第二相似度;A fourth calculation unit, configured to calculate a first similarity between the instant interest model and the search result information; calculate a second similarity between the long-term interest model and the search result information; 第四处理单元,用于根据所述第一相似度和第二相似度获取评分值,并根据所述评分值对所述搜索结果信息进行排序处理。The fourth processing unit is configured to acquire a score value according to the first similarity degree and the second similarity degree, and sort the search result information according to the score value. 29.一种搜索应用服务器,其特征在于,包括:29. A search application server, comprising: 第五接收模块,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;The fifth receiving module is configured to receive a search request message sent by a search client, where the search request message carries keyword information; 第五处理模块,用于从用户数据库中提取即时兴趣模型和长期兴趣模型;The fifth processing module is used to extract an immediate interest model and a long-term interest model from the user database; 第五发送模块,用于向搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型,以使所述搜索服务器根据所述即时兴趣模型和长期兴趣模型对所述关键词信息进行搜索。The fifth sending module is configured to send a search request to a search server, wherein the search request carries keyword information and the immediate interest model and the long-term interest model, so that the search server can search according to the immediate interest model and the long-term interest model The keyword information is searched. 30.根据权利要求29所述的搜索应用服务器,其特征在于,所述第五处理模块具体用于应用条件随机场模型计算在给定与所述搜索请求对应的当前查询q(t)处于同一搜索上下文会话的查询序列q(1),....q(t-1),q(t)的条件下,当前查询q(t)的输出类型的条件概率,将该条件概率值作为即时兴趣模型与该输出类型对应的兴趣维度的评分值。30. The search application server according to claim 29, characterized in that, the fifth processing module is specifically used to apply the conditional random field model to calculate the given current query q(t) corresponding to the search request at the same Under the condition of the query sequence q(1), ... q(t-1), q(t) of the search context session, the conditional probability of the output type of the current query q(t), the conditional probability value is used as the instant The scoring value of the interest dimension corresponding to the interest model and this output type. 31.一种移动搜索系统,其特征在于,包括:31. A mobile search system, comprising: 第一搜索应用服务器,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第一搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;The first search application server is configured to receive a search request message sent by a search client, wherein the search request message carries keyword information; extract an immediate interest model and a long-term interest model from a user database; send a search request to the first search server , the search request carries keyword information and the immediate interest model and long-term interest model; 第一搜索服务器,用于接收所述第一搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及所述第一搜索应用服务器获取的即时兴趣模型和长期兴趣模型;根据所述搜索请求、各成员引擎的元索引信息以及所述即时兴趣模型和长期兴趣模型,计算所述成员引擎的相关度评分值,根据所述相关度评分值从第一成员引擎设备中选择一个或多个成员引擎对所述关键词信息进行搜索;The first search server is configured to receive the search request sent by the first search application server, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the first search application server ; According to the search request, the meta-index information of each member engine and the instant interest model and long-term interest model, calculate the relevance score value of the member engine, according to the correlation score value from the first member engine device Selecting one or more member engines to search for the keyword information; 第一成员引擎设备,用于接收所述第一搜索服务器发送的搜索请求,对所述关键词信息进行搜索,并将搜索结果信息发送给所述第一搜索服务器,以使所述第一搜索服务器通过所述第一搜索应用服务器将所述搜索结果信息反馈给所述搜索客户端。The first member engine device is configured to receive a search request sent by the first search server, search for the keyword information, and send search result information to the first search server, so that the first search The server feeds back the search result information to the search client through the first search application server. 32.一种移动搜索系统,其特征在于,包括:32. A mobile search system, comprising: 第二搜索应用服务器,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第二搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;The second search application server is configured to receive the search request message sent by the search client, the search request message carries keyword information; extract the immediate interest model and the long-term interest model from the user database; send a search request to the second search server , the search request carries keyword information and the immediate interest model and long-term interest model; 第二搜索服务器,用于向第二成员引擎设备发送搜索请求,所述搜索请求中携带所需搜索的关键词信息以及第二搜索应用服务器获取的即时兴趣模型和长期兴趣模型;接收所述第二成员引擎设备反馈的搜索结果信息以及与所述搜索结果信息对应的评分信息;根据所述评分信息和相关因素信息对所述搜索结果信息进行重新评分排序,获取重新评分排序后的搜索结果信息,并将所述重新评分排序后的搜索结果信息发送给所述第二搜索应用服务器;The second search server is configured to send a search request to the second member engine device, the search request carries keyword information to be searched and the immediate interest model and long-term interest model acquired by the second search application server; receiving the first The search result information fed back by the two-member engine device and the scoring information corresponding to the search result information; re-scoring and sorting the search result information according to the scoring information and related factor information, and obtaining the re-scoring and sorting search result information , and sending the re-scored and sorted search result information to the second search application server; 第二成员引擎设备,用于接收所述第二搜索服务器发送的搜索请求,对所述关键词信息进行搜索,根据所述即时兴趣模型和长期兴趣模型获取搜索结果信息以及与所述搜索结果信息对应的评分信息,并将所述搜索结果信息以及评分信息发送给所述第二搜索服务器。The second member engine device is configured to receive a search request sent by the second search server, search for the keyword information, obtain search result information and match the search result information according to the immediate interest model and the long-term interest model Corresponding scoring information, and sending the search result information and scoring information to the second search server. 33.一种移动搜索系统,其特征在于,包括:33. A mobile search system, comprising: 第三搜索应用服务器,用于接收搜索客户端发送的搜索请求消息,所述搜索请求消息中携带关键词信息;从用户数据库中提取即时兴趣模型和长期兴趣模型;向第三搜索服务器发送搜索请求,所述搜索请求中携带关键词信息以及所述即时兴趣模型和长期兴趣模型;The third search application server is configured to receive the search request message sent by the search client, the search request message carries keyword information; extract the immediate interest model and the long-term interest model from the user database; send a search request to the third search server , the search request carries keyword information and the immediate interest model and long-term interest model; 第三搜索服务器,用于接收第三搜索应用服务器发送的搜索请求,所述搜索请求中携带所需搜索的关键词信息以及搜索应用服务器获取的即时兴趣模型和长期兴趣模型;接收第三成员引擎设备根据所述关键词信息搜索获取的搜索结果信息,并根据所述即时兴趣模型和长期兴趣模型对所述搜索结果信息进行评分排序处理;并将评分排序处理后的搜索结果信息以及相应的评分信息发送给所述第三搜索应用服务器;The third search server is used to receive the search request sent by the third search application server, the search request carries the keyword information to be searched and the immediate interest model and long-term interest model obtained by the search application server; receiving the third member engine The device searches the search result information acquired according to the keyword information, and performs scoring and sorting processing on the search result information according to the immediate interest model and the long-term interest model; sorts the search result information after scoring and corresponding scoring sending the information to the third search application server; 第三成员引擎设备,用于接收所述第三搜索服务器发送的搜索请求,对所述关键词信息进行搜索,获取所述搜索结果信息。The third member engine device is configured to receive the search request sent by the third search server, search the keyword information, and obtain the search result information.
CN200910220810.7A 2009-11-06 2009-11-06 Mobile search method, device and system Active CN102056335B (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN200910220810.7A CN102056335B (en) 2009-11-06 2009-11-06 Mobile search method, device and system
PCT/CN2010/077773 WO2011054245A1 (en) 2009-11-06 2010-10-15 Mobile search method, device and system
US13/464,391 US8983971B2 (en) 2009-11-06 2012-05-04 Method, apparatus, and system for mobile search

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN200910220810.7A CN102056335B (en) 2009-11-06 2009-11-06 Mobile search method, device and system

Publications (2)

Publication Number Publication Date
CN102056335A true CN102056335A (en) 2011-05-11
CN102056335B CN102056335B (en) 2014-02-19

Family

ID=43960105

Family Applications (1)

Application Number Title Priority Date Filing Date
CN200910220810.7A Active CN102056335B (en) 2009-11-06 2009-11-06 Mobile search method, device and system

Country Status (3)

Country Link
US (1) US8983971B2 (en)
CN (1) CN102056335B (en)
WO (1) WO2011054245A1 (en)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102890689A (en) * 2011-07-22 2013-01-23 北京百度网讯科技有限公司 Method and system for building user interest model
CN102956014A (en) * 2011-08-12 2013-03-06 源初科技有限公司 Method of attention-targeting for online advertisement
CN102970352A (en) * 2012-11-08 2013-03-13 北京小米科技有限责任公司 Method and device for providing information
WO2014032244A1 (en) * 2012-08-30 2014-03-06 Microsoft Corporation Feature-based candidate selection
CN103744830A (en) * 2013-12-31 2014-04-23 北京锐安科技有限公司 Semantic analysis based identification method of identity information in EXCEL document
CN103870458A (en) * 2012-12-07 2014-06-18 富士通株式会社 Data processing device, data processing method and data processing program
CN104679743A (en) * 2013-11-26 2015-06-03 阿里巴巴集团控股有限公司 Method and device for determining preference model of user
CN104850537A (en) * 2014-02-17 2015-08-19 腾讯科技(深圳)有限公司 Method and device for screening text content
CN106415535A (en) * 2014-04-14 2017-02-15 微软技术许可有限责任公司 Context-sensitive search using a deep learning model
CN107291442A (en) * 2017-04-28 2017-10-24 中国农业大学 The choosing method and tourism game design method of a kind of game design model of travelling
US9921665B2 (en) 2012-06-25 2018-03-20 Microsoft Technology Licensing, Llc Input method editor application platform
CN104915429B (en) * 2015-06-15 2018-09-04 小米科技有限责任公司 Keyword search methodology and device
CN110245299A (en) * 2019-06-19 2019-09-17 中国人民解放军国防科技大学 A sequential recommendation method and system based on dynamic interactive attention mechanism
CN111078849A (en) * 2019-12-02 2020-04-28 百度在线网络技术(北京)有限公司 Method and apparatus for outputting information
CN112182387A (en) * 2020-09-29 2021-01-05 中国人民大学 A Personalized Search Algorithm Enhanced by Time Information

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030217106A1 (en) * 2002-03-25 2003-11-20 Eytan Adar System and method for profiling clients within a system for harvesting community knowledge
CN103377200B (en) * 2012-04-17 2018-09-04 腾讯科技(深圳)有限公司 User preference information acquisition method and device
US10210547B1 (en) 2013-11-25 2019-02-19 Groupon, Inc. Predictive recommendation system using contextual relevance
US8837835B1 (en) 2014-01-20 2014-09-16 Array Technology, LLC Document grouping system
US10409823B2 (en) * 2015-12-29 2019-09-10 Facebook, Inc. Identifying content for users on online social networks
US10037365B2 (en) * 2016-01-29 2018-07-31 Integral Search International Ltd. Computer-implemented patent searching method in connection to matching degree
CN109033140B (en) * 2018-06-08 2020-05-29 北京百度网讯科技有限公司 Method, device, equipment and computer storage medium for determining search result
JP7183600B2 (en) 2018-07-20 2022-12-06 株式会社リコー Information processing device, system, method and program

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6963867B2 (en) * 1999-12-08 2005-11-08 A9.Com, Inc. Search query processing to provide category-ranked presentation of search results
US6785671B1 (en) * 1999-12-08 2004-08-31 Amazon.Com, Inc. System and method for locating web-based product offerings
US7428529B2 (en) * 2004-04-15 2008-09-23 Microsoft Corporation Term suggestion for multi-sense query
US20070067305A1 (en) * 2005-09-21 2007-03-22 Stephen Ives Display of search results on mobile device browser with background process
CN101140582A (en) * 2007-09-27 2008-03-12 中兴通讯股份有限公司 User preferences generation system and method thereof

Cited By (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102890689A (en) * 2011-07-22 2013-01-23 北京百度网讯科技有限公司 Method and system for building user interest model
CN102956014A (en) * 2011-08-12 2013-03-06 源初科技有限公司 Method of attention-targeting for online advertisement
US10867131B2 (en) 2012-06-25 2020-12-15 Microsoft Technology Licensing Llc Input method editor application platform
US9921665B2 (en) 2012-06-25 2018-03-20 Microsoft Technology Licensing, Llc Input method editor application platform
WO2014032244A1 (en) * 2012-08-30 2014-03-06 Microsoft Corporation Feature-based candidate selection
CN104823183A (en) * 2012-08-30 2015-08-05 微软技术许可有限责任公司 Feature-based candidate selection
US9767156B2 (en) 2012-08-30 2017-09-19 Microsoft Technology Licensing, Llc Feature-based candidate selection
CN102970352A (en) * 2012-11-08 2013-03-13 北京小米科技有限责任公司 Method and device for providing information
CN103870458A (en) * 2012-12-07 2014-06-18 富士通株式会社 Data processing device, data processing method and data processing program
CN103870458B (en) * 2012-12-07 2017-07-18 富士通株式会社 Data processing equipment, data processing method and program
CN104679743A (en) * 2013-11-26 2015-06-03 阿里巴巴集团控股有限公司 Method and device for determining preference model of user
CN103744830A (en) * 2013-12-31 2014-04-23 北京锐安科技有限公司 Semantic analysis based identification method of identity information in EXCEL document
CN104850537A (en) * 2014-02-17 2015-08-19 腾讯科技(深圳)有限公司 Method and device for screening text content
CN104850537B (en) * 2014-02-17 2017-12-15 腾讯科技(深圳)有限公司 The method and device screened to content of text
CN106415535B (en) * 2014-04-14 2020-03-17 微软技术许可有限责任公司 Context-dependent search using deep learning models
CN106415535A (en) * 2014-04-14 2017-02-15 微软技术许可有限责任公司 Context-sensitive search using a deep learning model
CN104915429B (en) * 2015-06-15 2018-09-04 小米科技有限责任公司 Keyword search methodology and device
CN107291442B (en) * 2017-04-28 2020-01-10 中国农业大学 Method for selecting travel game design model and travel game design method
CN107291442A (en) * 2017-04-28 2017-10-24 中国农业大学 The choosing method and tourism game design method of a kind of game design model of travelling
CN110245299A (en) * 2019-06-19 2019-09-17 中国人民解放军国防科技大学 A sequential recommendation method and system based on dynamic interactive attention mechanism
CN110245299B (en) * 2019-06-19 2022-02-08 中国人民解放军国防科技大学 Sequence recommendation method and system based on dynamic interaction attention mechanism
CN111078849A (en) * 2019-12-02 2020-04-28 百度在线网络技术(北京)有限公司 Method and apparatus for outputting information
CN112182387A (en) * 2020-09-29 2021-01-05 中国人民大学 A Personalized Search Algorithm Enhanced by Time Information
CN112182387B (en) * 2020-09-29 2023-08-25 中国人民大学 A Time Information Enhanced Personalized Search Method

Also Published As

Publication number Publication date
CN102056335B (en) 2014-02-19
WO2011054245A1 (en) 2011-05-12
US8983971B2 (en) 2015-03-17
US20120221544A1 (en) 2012-08-30

Similar Documents

Publication Publication Date Title
CN102056335B (en) Mobile search method, device and system
CN101661474B (en) Search method and system
CN102760138B (en) Classification method and device for user network behaviors and search method and device for user network behaviors
CN108717407B (en) Entity vector determination method and device, and information retrieval method and device
CN103593425B (en) Intelligent retrieval method and system based on preference
CN108363821A (en) A kind of information-pushing method, device, terminal device and storage medium
CN110390094B (en) Method, electronic device and computer program product for classifying documents
CN103838756A (en) Method and device for determining pushed information
CN107357793A (en) Information recommendation method and device
CN103309886A (en) Trading-platform-based structural information searching method and device
CN112100396A (en) Data processing method and device
CN104199965A (en) Semantic information retrieval method
CN103164539A (en) Interactive type image retrieval method of combining user evaluation and labels
CN101685456B (en) A search method, system and device
WO2011087904A1 (en) Matching of advertising sources and keyword sets in online commerce platforms
WO2010096986A1 (en) Mobile search method and device
CN103744918A (en) Vertical domain based micro blog searching ranking method and system
CN115905489B (en) Method for providing bidding information search service
CN119782494A (en) Search optimization method, device and medium based on large model
KR20120087346A (en) System and method for providing information between coperations and customers
CN105975508A (en) Personalized meta-search engine searched result merging and sorting method
CN110008407A (en) Information retrieval method and device
CN114282536A (en) Intelligent recommendation engine system based on AI algorithm
CN112765311A (en) Method for searching referee document
CN112270199A (en) CGAN (Carrier-grade network Access network) method based personalized semantic space keyword Top-K query method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant